Hydroelectric generating set fault detection method and system based on deep learning

Abstract

The invention relates to a hydroelectric generating set fault detection method and system based on deep learning, and the method comprises the steps: forming a target vibration vector, a target throw vector and a target pulsation vector according to a current vibration vector set, a current throw vector set and a current pulsation vector set; and constructing a vibration vector three-dimensional trend graph, a throw vector three-dimensional trend graph and a pulsation vector three-dimensional trend graph, extracting feature monitoring data, performing fault judgment according to the feature monitoring data to obtain node judgment data, judging whether a fault exists according to the node judgment data, and if so, identifying a fault type, a fault level and a fault development prediction graph. The invention further provides a system for realizing fault detection of the hydroelectric generating set based on deep learning, electronic equipment and a computer readable storage medium. According to the invention, the problems of low fault analysis efficiency and poor early warning precision when fault detection is carried out on the hydroelectric generating set by using a deep learning algorithm at present can be solved.

Description

Technical Field

[0001] The present invention relates to the technical field of hydro-generator sets, and particularly to a fault detection method, system, electronic device and computer-readable storage medium for hydro-generator sets based on deep learning implementation. Background Art

[0002] A hydro-generator set is an electromechanical device that converts water energy into electrical energy. Due to the complex working environment, various equipment failures may occur in the hydro-generator set, such as flooding of the unit, generator stator rubbing, burning of generator bars, breakage of turbine blades, and explosion of transformers. Therefore, the monitoring and diagnosis of the operating state of the hydro-generator set are particularly important. During the operation of the hydro-generator set, various parameters such as temperature, vibration value, swing value, water pressure pulsation value, and stator-rotor air gap will show abnormal trends. Since vibration, swing, and water pressure pulsation signals contain a large amount of equipment operation information, they are the most basic and reliable monitoring indicators for hydro-generator sets.

[0003] Currently, the monitoring of hydro-generator sets mainly involves first obtaining the vibration signals of the upper frame, lower frame, and top cover, the swing signals of the upper guide, lower guide, and water guide, and the water pressure pulsation signals of the spiral case, top cover, and draft tube, and then analyzing the operating state based on the vibration signals, swing signals, and water pressure pulsation signals. For example, using deep learning algorithms to detect faults in hydro-generator sets based on vibration signals, swing signals, and water pressure pulsation signals. However, the current fault detection of hydro-generator sets using deep learning algorithms only simply analyzes the original monitoring data, does not perform structured processing on the original monitoring data, and cannot predict the fault development trend based on the original monitoring data, making it difficult to carry out refined early warning. Therefore, the current fault detection of hydro-generator sets using deep learning algorithms has problems of low fault analysis efficiency and poor early warning accuracy. Summary of the Invention

[0004] The present invention provides a fault detection method, system and computer-readable storage medium for hydro-generator sets based on deep learning implementation, and its main purpose is to solve the problems of low fault analysis efficiency and poor early warning accuracy existing in the current fault detection of hydro-generator sets using deep learning algorithms.

[0005] To achieve the above object, a fault detection method for hydro-generator sets based on deep learning implementation provided by the present invention includes:

[0006] Obtain the real-time monitoring trend atlas of the hydro-generator set, where the real-time monitoring trend atlas includes the upper frame vibration trend graph, lower frame vibration trend graph, top cover vibration trend graph, upper guide swing trend graph, lower guide swing trend graph, water guide swing trend graph, spiral case water pressure pulsation trend graph, top cover water pressure pulsation trend graph, and draft tube water pressure pulsation trend graph;

[0007] Identify the current vibration vector sets of the upper frame vibration trend chart, the lower frame vibration trend chart, and the top cover vibration trend chart, and synthesize a target vibration vector in a pre-constructed vibration three-dimensional coordinate system according to a pre-constructed vector synthesis formula, where the vector synthesis formula is as follows:

[0008]

[0009] Among them, represents the target vibration vector, represents the current vibration vector of the upper frame in the current vibration vector set, represents the current vibration vector of the lower frame in the current vibration vector set, represents the current vibration vector of the top cover in the current vibration vector set;

[0010] Identify the current swing vector sets of the upper guide swing trend chart, the lower guide swing trend chart, and the water guide swing trend chart, and synthesize a target swing vector in the swing three-dimensional coordinate system according to the current swing vector sets;

[0011] Identify the current pulsation vector sets of the volute water pressure pulsation trend chart, the top cover water pressure pulsation trend chart, and the draft tube water pressure pulsation trend chart, and synthesize a target pulsation vector in the pulsation three-dimensional coordinate system according to the current pulsation vector sets;

[0012] Construct a three-dimensional trend chart of vibration vectors, a three-dimensional trend chart of swing vectors, and a three-dimensional trend chart of pulsation vectors respectively according to the target vibration vector, the target swing vector, and the target pulsation vector;

[0013] Extract characteristic monitoring data from the three-dimensional trend chart of vibration vectors, the three-dimensional trend chart of swing vectors, and the three-dimensional trend chart of pulsation vectors in sequence;

[0014] Input the characteristic monitoring data into a pre-constructed target neural network for fault determination to obtain node determination data;

[0015] Judge whether the hydropower unit has a fault according to the node determination data;

[0016] If the hydropower unit has no fault, return to the above steps of obtaining the real-time monitoring trend atlas of the hydropower unit;

[0017] If the hydropower unit has a fault, identify the fault type and fault level of the hydropower unit according to the node determination data and conduct an operation trend prediction to obtain a fault development prediction diagram, completing the fault detection of the hydropower unit based on deep learning.

[0018] Optionally, the identification of the current vibration vector sets of the upper frame vibration trend chart, the lower frame vibration trend chart, and the top cover vibration trend chart includes:

[0019] Identify the current monitoring signal point of the upper frame in the vibration trend diagram of the upper frame;

[0020] According to the preset vibration sampling period and the current monitoring signal point of the upper frame, extract the adjacent monitoring signal points of the upper frame in the vibration trend diagram of the upper frame;

[0021] Construct the current vibration vector of the upper frame with the adjacent monitoring signal point of the upper frame as the starting point and the current monitoring signal point of the upper frame as the ending point;

[0022] Identify the current monitoring signal point of the lower frame in the vibration trend diagram of the lower frame;

[0023] According to the vibration sampling period and the current monitoring signal point of the lower frame, extract the adjacent monitoring signal points of the lower frame in the vibration trend diagram of the lower frame;

[0024] Construct the current vibration vector of the lower frame with the adjacent monitoring signal point of the lower frame as the starting point and the current monitoring signal point of the lower frame as the ending point;

[0025] Identify the current monitoring signal point of the top cover in the vibration trend diagram of the top cover;

[0026] According to the vibration sampling period and the current monitoring signal point of the top cover, extract the adjacent monitoring signal points of the top cover in the vibration trend diagram of the top cover;

[0027] Construct the current vibration vector of the top cover with the adjacent monitoring signal point of the top cover as the starting point and the current monitoring signal point of the top cover as the ending point;

[0028] Summarize the current vibration vector of the upper frame, the current vibration vector of the lower frame, and the current vibration vector of the top cover to obtain the current vibration vector set.

[0029] Optionally, the synthesizing the target vibration vector in the pre-constructed vibration three-dimensional coordinate system according to the pre-constructed vector synthesis formula and the current vibration vector set includes:

[0030] Extract the current vibration vector of the upper frame from the current vibration vector set, and identify the upper frame coordinate plane of the vibration three-dimensional coordinate system;

[0031] Fill the current vibration vector of the upper frame into the upper frame coordinate plane to obtain the upper frame vector to be synthesized, and the starting point of the upper frame vector to be synthesized coincides with the origin of the vibration three-dimensional coordinate system;

[0032] Extract the current vibration vector of the lower frame from the current vibration vector set, and identify the lower frame coordinate plane of the vibration three-dimensional coordinate system;

[0033] Fill the current vibration vector of the lower frame into the coordinate plane of the lower frame to obtain a vector to be synthesized for the lower frame, and the starting point of the vector to be synthesized for the lower frame coincides with the origin of the vibration three-dimensional coordinate system;

[0034] Extract the current vibration vector of the top cover from the current vibration vector set, and identify the coordinate plane of the top cover in the vibration three-dimensional coordinate system;

[0035] Fill the current vibration vector of the top cover into the coordinate plane of the top cover to obtain a vector to be synthesized for the top cover, and the starting point of the vector to be synthesized for the top cover coincides with the origin of the vibration three-dimensional coordinate system;

[0036] According to the vector to be synthesized for the upper frame, the vector to be synthesized for the lower frame, and the vector to be synthesized for the top cover, use the vector synthesis formula to synthesize a target vibration vector in the vibration three-dimensional coordinate system.

[0037] Optionally, the constructing the three-dimensional trend diagrams of the vibration vector, the swing vector, and the pulsation vector respectively according to the target vibration vector, the target swing vector, and the target pulsation vector includes:

[0038] Obtain a historical three-dimensional trend diagram of vibration, and determine whether there is a preset vibration connection vector in the historical three-dimensional trend diagram of vibration, where the vibration connection vector refers to a vibration vector in the historical three-dimensional trend diagram of vibration that can be connected to the target vibration vector;

[0039] If there is no vibration connection vector in the historical three-dimensional trend diagram of vibration, translate the starting point of the target vibration vector to the coordinate origin of the historical three-dimensional trend diagram of vibration to obtain a three-dimensional trend diagram of the vibration vector;

[0040] If there is a vibration connection vector in the historical three-dimensional trend diagram of vibration, identify the termination point of the vibration connection vector;

[0041] Connect the target vibration vector to the termination point of the vibration connection vector to obtain a three-dimensional trend diagram of the vibration vector;

[0042] Obtain a historical three-dimensional trend diagram of swing, and determine whether there is a preset swing connection vector in the historical three-dimensional trend diagram of swing, where the swing connection vector refers to a swing vector in the historical three-dimensional trend diagram of swing that can be connected to the target swing vector;

[0043] If there is no swing connection vector in the historical three-dimensional trend diagram of swing, translate the starting point of the target swing vector to the coordinate origin of the historical three-dimensional trend diagram of swing to obtain a three-dimensional trend diagram of the swing vector;

[0044] If there is a swing connection vector in the historical three-dimensional trend diagram of swing, identify the termination point of the swing connection vector;

[0045] Connect the target deflection vector to the end point of the deflection connection vector to obtain a three-dimensional trend graph of the deflection vector;

[0046] Obtain a historical pulsation three-dimensional trend graph, and determine whether there is a preset pulsation connection vector in the historical pulsation three-dimensional trend graph, where the pulsation connection vector refers to a pulsation vector in the historical vibration three-dimensional trend graph that can be connected to the target pulsation vector;

[0047] If there is no pulsation connection vector in the historical pulsation three-dimensional trend graph, translate the starting point of the target pulsation vector to the coordinate origin of the historical pulsation three-dimensional trend graph to obtain a three-dimensional trend graph of the pulsation vector;

[0048] If there is a pulsation connection vector in the historical pulsation three-dimensional trend graph, identify the end point of the pulsation connection vector;

[0049] Connect the target pulsation vector to the end point of the pulsation connection vector to obtain a three-dimensional trend graph of the pulsation vector.

[0050] Optionally, the determining whether there is a preset vibration connection vector in the historical vibration three-dimensional trend graph includes:

[0051] Obtain the current monitoring time, and construct a connection vector monitoring period according to the vibration sampling period using the following formula:

[0052] X j =[t n -T,t n

[0053] where X j represents the connection vector monitoring period, t n represents the current monitoring time, and T represents the vibration sampling period;

[0054] Determine whether there is a vibration vector in the historical vibration three-dimensional trend graph during the connection vector monitoring period;

[0055] If there is no vibration vector in the historical vibration three-dimensional trend graph during the connection vector monitoring period, then there is no vibration connection vector in the historical vibration three-dimensional trend graph;

[0056] If there is a vibration vector in the historical vibration three-dimensional trend graph during the connection vector monitoring period, then there is a vibration connection vector in the historical vibration three-dimensional trend graph.

[0057] Optionally, the sequentially extracting characteristic monitoring data from the vibration vector three-dimensional trend graph, the deflection vector three-dimensional trend graph, and the pulsation vector three-dimensional trend graph includes:

[0058] ​According to the preset vibration sampling value, swing sampling value, and pulsation sampling value respectively, collect the vibration turning point set, swing turning point set, and pulsation turning point set in the three-dimensional trend diagram of the vibration vector, the three-dimensional trend diagram of the swing vector, and the three-dimensional trend diagram of the pulsation vector;

[0059] Identify the vibration sample point coordinates of each collected vibration turning point in the collected vibration turning point set to obtain a vibration coordinate set, identify the swing sample point coordinates of each swing turning point in the swing turning point set to obtain a swing coordinate set, and identify the pulsation sample point coordinates of each pulsation turning point in the pulsation turning point set to obtain a pulsation coordinate set;

[0060] Use the vibration coordinate set, swing coordinate set, and pulsation coordinate set as feature monitoring data.

[0061] Optionally, before inputting the feature monitoring data into a pre-constructed target neural network for fault determination to obtain node determination data, the method further includes:

[0062] Successively extract the vibration training trend diagram, swing training trend diagram, and pulsation training trend diagram from a pre-constructed standard training trend diagram set;

[0063] According to the vibration sampling value, swing sampling value, and pulsation sampling value, respectively extract the vibration training coordinate set, swing training coordinate set, and pulsation training coordinate set in segments from the vibration training trend diagram, swing training trend diagram, and pulsation training trend diagram;

[0064] Use the vibration training coordinate set, swing training coordinate set, and pulsation training coordinate set to train a pre-constructed original neural network to obtain iterative output data, where the original neural network includes an input layer, a hidden layer, and an output layer, and the input layer further includes vibration coordinate input node 1, vibration coordinate input node 2,..., vibration coordinate input node z, swing coordinate input node 1, swing coordinate input node 2,..., swing coordinate input node b, pulsation coordinate input node 1, pulsation coordinate input node 2,..., pulsation coordinate input node m, and the output layer further includes a fault type output node, a fault level output node, a first time development level prediction node, a second time development level prediction node,..., an fth time development level prediction node;

[0065] Obtain the fault type, fault level, first time fault development level, second time fault development level,..., fth time fault development level of the vibration training trend diagram, swing training trend diagram, and pulsation training trend diagram;

[0066] Based on the iterative output data, the original neural network is iteratively adjusted in parameters by using the fault type, fault level, fault development level at the first time, fault development level at the second time, …, fault development level at the f-th time to obtain a target neural network.

[0067] Optionally, the identifying the fault type and fault level of the hydropower unit according to the node determination data and performing an operation trend prediction to obtain a fault development prediction diagram includes:

[0068] Sequentially extracting the fault type output value, fault level output value, first-time development prediction value, second-time development prediction value, …, f-th time development prediction value of the fault type output node, fault level output node, first-time development level prediction node, second-time development level prediction node, …, f-th time development level prediction node in the node determination data;

[0069] Determining the fault type of the hydropower unit according to the fault type output value, determining the fault level of the hydropower unit according to the fault level output value, and determining the first-time fault development level, second-time fault development level, …, f-th time fault development level of the hydropower unit according to the first-time development prediction value, second-time development prediction value, …, f-th time development prediction value;

[0070] Drawing the fault development prediction diagram according to the first-time fault development level, second-time fault development level, …, f-th time fault development level.

[0071] Optionally, after the identifying the fault type and fault level of the hydropower unit according to the node determination data and performing an operation trend prediction to obtain a fault development prediction diagram, the method further includes:

[0072] Obtaining the current time and judging the current fault development level of the current time in the fault development prediction diagram;

[0073] Identifying the current warning level corresponding to the current fault development level and performing a fault warning according to the current warning level.

[0074] To solve the above problems, the present invention further provides a fault detection system for a hydropower unit based on deep learning, and the system includes:

[0075] A real-time monitoring trend atlas acquisition module, configured to acquire a real-time monitoring trend atlas of the hydropower unit, where the real-time monitoring trend atlas includes an upper frame vibration trend diagram, a lower frame vibration trend diagram, a top cover vibration trend diagram, an upper guide swing trend diagram, a lower guide swing trend diagram, a water guide swing trend diagram, a spiral case water pressure pulsation trend diagram, a top cover water pressure pulsation trend diagram, and a draft tube water pressure pulsation trend diagram;

[0076] A target vector synthesis module, configured to identify the current vibration vector sets of the upper frame vibration trend graph, the lower frame vibration trend graph, and the top cover vibration trend graph, and synthesize a target vibration vector in a pre-constructed vibration three-dimensional coordinate system according to a pre-constructed vector synthesis formula and the current vibration vector sets, where the vector synthesis formula is as follows:

[0077]

[0078] Wherein, represents the target vibration vector, represents the current vibration vector of the upper frame in the current vibration vector set, represents the current vibration vector of the lower frame in the current vibration vector set, represents the current vibration vector of the top cover in the current vibration vector set; identify the current swing vector sets of the upper guide swing trend graph, the lower guide swing trend graph, and the water guide swing trend graph, and synthesize a target swing vector in the swing three-dimensional coordinate system according to the current swing vector sets; identify the current pulsation vector sets of the spiral case water pressure pulsation trend graph, the top cover water pressure pulsation trend graph, and the draft tube water pressure pulsation trend graph, and synthesize a target pulsation vector in the pulsation three-dimensional coordinate system;

[0079] A vector three-dimensional trend graph construction module, configured to construct a vibration vector three-dimensional trend graph, a swing vector three-dimensional trend graph, and a pulsation vector three-dimensional trend graph respectively according to the target vibration vector, the target swing vector, and the target pulsation vector;

[0080] A feature monitoring data extraction module, configured to sequentially extract feature monitoring data from the vibration vector three-dimensional trend graph, the swing vector three-dimensional trend graph, and the pulsation vector three-dimensional trend graph;

[0081] A fault determination and prediction module, configured to input the feature monitoring data into a pre-constructed target neural network for fault determination to obtain node determination data; judge whether the hydropower unit has a fault according to the node determination data; if the hydropower unit has no fault, return to the above step of obtaining the real-time monitoring trend atlas of the hydropower unit; if the hydropower unit has a fault, identify the fault type and fault level of the hydropower unit according to the node determination data and perform an operation trend prediction to obtain a fault development prediction graph.

[0082] To solve the above problems, the present invention also provides an electronic device, which includes:

[0083] At least one processor; and,

[0084] A memory communicatively connected to the at least one processor; wherein,

[0085] The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to implement the above-mentioned fault detection method for hydropower generating units based on deep learning.

[0086] To solve the above problems, the present invention also provides a computer-readable storage medium, in which at least one instruction is stored, and the at least one instruction is executed by a processor in an electronic device to implement the above-mentioned fault detection method for hydropower generating units based on deep learning.

[0087] In the embodiment of the present invention, it is first necessary to obtain the real-time monitoring trend atlas of the hydropower unit, and the real-time monitoring trend atlas includes the upper frame vibration trend diagram, the lower frame vibration trend diagram, the top cover vibration trend diagram, the upper guide swing trend diagram, the lower guide swing trend diagram, the water guide swing trend diagram, the spiral case water pressure pulsation trend diagram, the top cover water pressure pulsation trend diagram, and the draft tube water pressure pulsation trend diagram. Since it is necessary to monitor the hydropower unit in real time, it is necessary to identify the current vibration vector sets of the upper frame vibration trend diagram, the lower frame vibration trend diagram, and the top cover vibration trend diagram. The current vibration vector sets contain the real-time change information of the operating state of the hydropower unit. At this time, the target vibration vector can be synthesized in the pre-constructed vibration three-dimensional coordinate system according to the pre-constructed vector synthesis formula and the current vibration vector sets, so as to achieve the goal of structurally processing the current vibration vector. Similarly, it is necessary to identify the current swing vector sets of the upper guide swing trend diagram, the lower guide swing trend diagram, and the water guide swing trend diagram, synthesize the target swing vector in the swing three-dimensional coordinate system according to the current swing vector sets, and identify the current pulsation vector sets of the spiral case water pressure pulsation trend diagram, the top cover water pressure pulsation trend diagram, and the draft tube water pressure pulsation trend diagram, and then synthesize the target pulsation vector in the pulsation three-dimensional coordinate system according to the current pulsation vector sets. Since streamlining the real-time monitoring trend atlas can reduce the complexity of the target neural network during training, the target vibration vector, the target swing vector, and the target pulsation vector can be further streamlined. At this time, the three-dimensional vibration vector trend diagram, the three-dimensional swing vector trend diagram, and the three-dimensional pulsation vector trend diagram can be constructed according to the target vibration vector, the target swing vector, and the target pulsation vector respectively. Finally, the characteristic monitoring data can be extracted from the three-dimensional vibration vector trend diagram, the three-dimensional swing vector trend diagram, and the three-dimensional pulsation vector trend diagram. After obtaining the characteristic monitoring data, the characteristic monitoring data can be input into the pre-constructed target neural network for fault determination to obtain the node determination data, and then it can be judged whether the hydropower unit has a fault according to the node determination data. If the hydropower unit has a fault, the fault type and fault level of the hydropower unit can be further identified according to the node determination data and the operation trend can be predicted to obtain the fault development prediction diagram, so as to complete the fault detection of the hydropower unit based on deep learning. Therefore, the hydropower unit fault detection method, system, electronic device, and computer-readable storage medium based on deep learning proposed by the present invention can solve the problems of low fault analysis efficiency and poor warning accuracy in the current fault detection of hydropower units using deep learning algorithms. BRIEF DESCRIPTION OF THE DRAWINGS

[0088] Figure 1 FIG. is a schematic flow chart of a hydropower unit fault detection method based on deep learning provided by an embodiment of the present invention;

[0089] Figure 2Function module diagram of the hydro-generator unit fault detection system implemented based on deep learning provided by an embodiment of the present invention;

[0090] Figure 3 Structural schematic diagram of an electronic device for implementing the hydro-generator unit fault detection method implemented based on deep learning provided by an embodiment of the present invention.

[0091] The implementation of the object, functional features and advantages of the present invention will be further described with reference to the embodiments and the accompanying drawings. Specific embodiments

[0092] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0093] An embodiment of the present application provides a hydro-generator unit fault detection method implemented based on deep learning. The execution subject of the hydro-generator unit fault detection method implemented based on deep learning includes, but is not limited to, at least one of electronic devices such as a server, a terminal, etc. that can be configured to execute the method provided by the embodiment of the present application. In other words, the hydro-generator unit fault detection method implemented based on deep learning can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc.

[0094] Embodiment 1:

[0095] Refer to Figure 1 As shown, it is a flow schematic diagram of the hydro-generator unit fault detection method implemented based on deep learning provided by an embodiment of the present invention. In this embodiment, the hydro-generator unit fault detection method implemented based on deep learning includes:

[0096] S1. Obtain a real-time monitoring trend atlas of the hydro-generator unit, where the real-time monitoring trend atlas includes an upper frame vibration trend diagram, a lower frame vibration trend diagram, a top cover vibration trend diagram, an upper guide swing trend diagram, a lower guide swing trend diagram, a water guide swing trend diagram, a spiral case water pressure pulsation trend diagram, a top cover water pressure pulsation trend diagram, and a draft tube water pressure pulsation trend diagram.

[0097] It is understandable that the upper frame vibration trend diagram refers to a diagram showing the change trend of the vibration amplitude of the upper frame of the hydro-generator unit over time, the vertical coordinate can represent the amplitude, and the horizontal coordinate can represent the time. The lower frame vibration trend diagram refers to a diagram showing the change trend of the vibration amplitude of the lower frame of the hydro-generator unit over time, and the top cover vibration trend diagram refers to a diagram showing the change trend of the vibration amplitude of the top cover of the hydro-generator unit over time.

[0098] Further, the upper guide swing trend graph refers to the graph of the change trend of the swing of the upper guide bearing of the hydropower unit over time, the lower guide swing trend graph refers to the graph of the change trend of the swing of the lower guide bearing of the hydropower unit over time, and the water guide swing trend graph refers to the graph of the change trend of the swing of the water guide bearing of the hydropower unit over time.

[0099] It can be understood that the spiral case water pressure pulsation trend graph refers to the graph of the change trend of the water pressure pulse signal in the spiral case of the hydropower unit over time, the top cover water pressure pulsation trend graph refers to the graph of the change trend of the water pressure pulse signal in the top cover of the hydropower unit over time, and the draft tube water pressure pulsation trend graph refers to the graph of the change trend of the water pressure pulse signal in the draft tube of the hydropower unit over time.

[0100] It should be understood that in addition to the above-mentioned various trend graphs, the monitoring indicators of the hydropower unit also include indicators such as the stator-rotor air gap, temperature, and generator stator vibration data. However, the above-mentioned various trend graphs cover the most basic operation data of the hydropower unit. Therefore, monitoring indicators can be added according to the actual situation.

[0101] S2. Identify the current vibration vector sets of the upper frame vibration trend graph, the lower frame vibration trend graph, and the top cover vibration trend graph, and synthesize the target vibration vector in the pre-constructed vibration three-dimensional coordinate system according to the pre-constructed vector synthesis formula and the current vibration vector sets.

[0102] It can be understood that the current vibration vector sets include the upper frame current vibration vector, the lower frame current vibration vector, and the top cover current vibration vector. The upper frame current vibration vector refers to the vibration vector of the frame vibration trend graph at the current moment. For example, if the current moment is 7:00:00.01 seconds and the sampling frequency is 100 Hz, then the vibration sampling period is 0.01 s. When the amplitude corresponding to 7 o'clock in the upper frame vibration trend graph is 1250 mV and the amplitude at 7:00:00.01 seconds is 1252 mV, then the starting point of the upper frame current vibration vector is (7:00:00, 1250), the ending point is (7:00:01, 1252), and the vector direction points from the starting point to the ending point. The lower frame current vibration vector and the top cover current vibration vector are the same and will not be elaborated here.

[0103] It should be understood that the vector synthesis formula is as follows:

[0104]

[0105] Among them, represents the target vibration vector, represents the upper frame current vibration vector in the current vibration vector set, represents the lower frame current vibration vector in the current vibration vector set, represents the top cover current vibration vector in the current vibration vector set.

[0106] Furthermore, the vibration three-dimensional coordinate system includes three mutually perpendicular coordinate planes, namely the upper frame coordinate plane, the lower frame coordinate plane, and the top cover coordinate plane. The upper frame coordinate plane, the lower frame coordinate plane, and the top cover coordinate plane are respectively used to draw the current vibration vectors of the upper frame, the lower frame, and the top cover. The point where the upper frame coordinate plane, the lower frame coordinate plane, and the top cover coordinate plane intersect is the origin of the vibration three-dimensional coordinate system.

[0107] In the embodiment of the present invention, the current vibration vector set for identifying the vibration trend diagrams of the upper frame, the lower frame, and the top cover includes:

[0108] Identifying the current monitoring signal points of the upper frame in the vibration trend diagram of the upper frame;

[0109] According to the preset vibration sampling period and the current monitoring signal points of the upper frame, extracting the adjacent monitoring signal points of the upper frame in the vibration trend diagram of the upper frame;

[0110] Constructing the current vibration vector of the upper frame with the adjacent monitoring signal points of the upper frame as the starting point and the current monitoring signal points of the upper frame as the ending point;

[0111] Identifying the current monitoring signal points of the lower frame in the vibration trend diagram of the lower frame;

[0112] According to the vibration sampling period and the current monitoring signal points of the lower frame, extracting the adjacent monitoring signal points of the lower frame in the vibration trend diagram of the lower frame;

[0113] Constructing the current vibration vector of the lower frame with the adjacent monitoring signal points of the lower frame as the starting point and the current monitoring signal points of the lower frame as the ending point;

[0114] Identifying the current monitoring signal points of the top cover in the vibration trend diagram of the top cover;

[0115] According to the vibration sampling period and the current monitoring signal points of the top cover, extracting the adjacent monitoring signal points of the top cover in the vibration trend diagram of the top cover;

[0116] Constructing the current vibration vector of the top cover with the adjacent monitoring signal points of the top cover as the starting point and the current monitoring signal points of the top cover as the ending point;

[0117] Summarizing the current vibration vector of the upper frame, the current vibration vector of the lower frame, and the current vibration vector of the top cover to obtain the current vibration vector set.

[0118] Interpretable, the current monitoring signal point of the upper machine rack refers to the coordinate point of the vibration trend graph of the upper machine rack at the current moment. The vibration sampling period refers to the sampling time interval of the vibration trend graphs of the upper machine rack, the lower machine rack, and the top cover vibration trend graph. For example, when the sampling frequency is 100 Hz, the vibration sampling period is 0.01 s. The adjacent monitoring signal point of the upper machine rack refers to the coordinate point in the vibration trend graph of the upper machine rack that is closest to the current monitoring signal point of the upper machine rack. For example, when the current monitoring signal point of the upper machine rack is (7:00:01, 1252), under the condition that the vibration sampling period is 0.01 s, the adjacent monitoring signal point of the upper machine rack is the coordinate point corresponding to 7 o'clock.

[0119] Furthermore, the relevant noun explanations of the vibration trend graphs of the lower machine rack and the top cover vibration trend graph are the same as those of the vibration trend graph of the upper machine rack, and will not be elaborated here.

[0120] In the embodiment of the present invention, the synthesizing of the target vibration vector in the pre-constructed vibration three-dimensional coordinate system according to the pre-constructed vector synthesis formula and the current vibration vector set includes:

[0121] Extract the current vibration vector of the upper machine rack from the current vibration vector set, and identify the upper machine rack coordinate plane of the vibration three-dimensional coordinate system;

[0122] Fill the current vibration vector of the upper machine rack into the upper machine rack coordinate plane to obtain the to-be-synthesized vector of the upper machine rack, and the starting point of the to-be-synthesized vector of the upper machine rack coincides with the origin of the vibration three-dimensional coordinate system;

[0123] Extract the current vibration vector of the lower machine rack from the current vibration vector set, and identify the lower machine rack coordinate plane of the vibration three-dimensional coordinate system;

[0124] Fill the current vibration vector of the lower machine rack into the lower machine rack coordinate plane to obtain the to-be-synthesized vector of the lower machine rack, and the starting point of the to-be-synthesized vector of the lower machine rack coincides with the origin of the vibration three-dimensional coordinate system;

[0125] Extract the current vibration vector of the top cover from the current vibration vector set, and identify the top cover coordinate plane of the vibration three-dimensional coordinate system;

[0126] Fill the current vibration vector of the top cover into the top cover coordinate plane to obtain the to-be-synthesized vector of the top cover, and the starting point of the to-be-synthesized vector of the top cover coincides with the origin of the vibration three-dimensional coordinate system;

[0127] According to the to-be-synthesized vector of the upper machine rack, the to-be-synthesized vector of the lower machine rack, and the to-be-synthesized vector of the top cover, use the vector synthesis formula to synthesize the target vibration vector in the vibration three-dimensional coordinate system.

[0128] For example, when the end points of the upper frame vector to be synthesized, the lower frame vector to be synthesized, and the top cover vector to be synthesized are (0, 1), (0, 1), and (0, 1) respectively, the end point of the target vibration vector is (1, 1, 1).

[0129] S3. Identify the current set of swing vectors of the upper guide swing trend chart, the lower guide swing trend chart, and the water guide swing trend chart, and synthesize a target swing vector in the swing three-dimensional coordinate system according to the current set of swing vectors.

[0130] It can be understood that the acquisition methods and definitions of the current set of swing vectors, the swing three-dimensional coordinate system, and the target swing vector are the same as those of the current set of swing vectors, the vibration three-dimensional coordinate system, and the target vibration vector respectively, and will not be elaborated here.

[0131] S4. Identify the current set of pulsation vectors of the spiral case water pressure pulsation trend chart, the top cover water pressure pulsation trend chart, and the draft tube water pressure pulsation trend chart, and synthesize a target pulsation vector in the pulsation three-dimensional coordinate system according to the current set of pulsation vectors.

[0132] Furthermore, the acquisition methods and definitions of the current set of pulsation vectors, the pulsation three-dimensional coordinate system, and the target pulsation vector are the same as those of the current set of swing vectors, the vibration three-dimensional coordinate system, and the target vibration vector respectively, and will not be elaborated here.

[0133] S5. Construct a three-dimensional trend chart of vibration vectors, a three-dimensional trend chart of swing vectors, and a three-dimensional trend chart of pulsation vectors according to the target vibration vector, the target swing vector, and the target pulsation vector respectively.

[0134] It can be understood that the three-dimensional trend chart of vibration vectors, the three-dimensional trend chart of swing vectors, and the three-dimensional trend chart of pulsation vectors respectively refer to the three-dimensional trend charts obtained by connecting the heads and tails of multiple target vibration vectors, target swing vectors, and target pulsation vectors up to the current moment. For example: when the end point of the first target vibration vector is (1, 1, 1), the end point of the second target vibration vector is (1, 2, 1), and the end point of the third target vibration vector is (2, 3, 4), then the three-dimensional trend chart of vibration vectors can be obtained by connecting the heads and tails of the first, second, and third target vibration vectors in sequence, that is, translating the starting point of the second target vibration vector to the end point of the first target vibration vector, and translating the starting point of the third target vibration vector to the end point of the second target vibration vector. The three-dimensional trend chart of swing vectors and the three-dimensional trend chart of pulsation vectors are the same, and will not be elaborated here. The specific values of the target vibration vectors need to be set according to the actual situation.

[0135] In an embodiment of the present invention, the constructing a three-dimensional trend graph of vibration vectors, a three-dimensional trend graph of deflection vectors, and a three-dimensional trend graph of pulsation vectors respectively according to the target vibration vector, the target deflection vector, and the target pulsation vector includes:

[0136] Obtain a historical three-dimensional trend graph of vibration, and determine whether there is a preset vibration connection vector in the historical three-dimensional trend graph of vibration, where the vibration connection vector refers to a vibration vector in the historical three-dimensional trend graph of vibration that can be connected to the target vibration vector;

[0137] If there is no vibration connection vector in the historical three-dimensional trend graph of vibration, translate the starting point of the target vibration vector to the coordinate origin of the historical three-dimensional trend graph of vibration to obtain a three-dimensional trend graph of vibration vectors;

[0138] If there is a vibration connection vector in the historical three-dimensional trend graph of vibration, identify the termination point of the vibration connection vector;

[0139] Connect the target vibration vector to the termination point of the vibration connection vector to obtain a three-dimensional trend graph of vibration vectors;

[0140] Obtain a historical three-dimensional trend graph of deflection, and determine whether there is a preset deflection connection vector in the historical three-dimensional trend graph of deflection, where the deflection connection vector refers to a deflection vector in the historical three-dimensional trend graph of deflection that can be connected to the target deflection vector;

[0141] If there is no deflection connection vector in the historical three-dimensional trend graph of deflection, translate the starting point of the target deflection vector to the coordinate origin of the historical three-dimensional trend graph of deflection to obtain a three-dimensional trend graph of deflection vectors;

[0142] If there is a deflection connection vector in the historical three-dimensional trend graph of deflection, identify the termination point of the deflection connection vector;

[0143] Connect the target deflection vector to the termination point of the deflection connection vector to obtain a three-dimensional trend graph of deflection vectors;

[0144] Obtain a historical three-dimensional trend graph of pulsation, and determine whether there is a preset pulsation connection vector in the historical three-dimensional trend graph of pulsation, where the pulsation connection vector refers to a pulsation vector in the historical three-dimensional trend graph of vibration that can be connected to the target pulsation vector;

[0145] If there is no pulsation connection vector in the historical three-dimensional trend graph of pulsation, translate the starting point of the target pulsation vector to the coordinate origin of the historical three-dimensional trend graph of pulsation to obtain a three-dimensional trend graph of pulsation vectors;

[0146] If there is a pulsation connection vector in the historical three-dimensional trend graph of pulsation, identify the termination point of the pulsation connection vector;

[0147] Connect the target pulsation vector to the termination point of the pulsation connection vector to obtain a three-dimensional trend graph of the pulsation vector.

[0148] It can be understood that the three-dimensional trend graph of the historical vibration refers to the three-dimensional trend graph of vibration obtained by connecting each target vibration vector before the current moment. The vibration vector that can be connected to the target vibration vector refers to the vibration vector whose sampling moment at the termination point of the three-dimensional trend graph of the historical vibration is the same as the sampling moment at the starting point of the target vibration vector. The definitions and related noun explanations of the three-dimensional trend graph of the swing vector and the three-dimensional trend graph of the pulsation vector refer to the three-dimensional trend graph of the vibration vector, which will not be elaborated here.

[0149] In an embodiment of the present invention, determining whether there is a preset vibration connection vector in the three-dimensional trend graph of the historical vibration includes:

[0150] Obtain the current monitoring moment, and construct a connection vector monitoring period according to the vibration sampling period using the following formula:

[0151] X j =[t n -T, t n

[0152] where X j represents the connection vector monitoring period, t n represents the current monitoring moment, and T represents the vibration sampling period;

[0153] Determine whether there is a vibration vector in the three-dimensional trend graph of the historical vibration during the connection vector monitoring period;

[0154] If there is no vibration vector in the three-dimensional trend graph of the historical vibration during the connection vector monitoring period, then there is no vibration connection vector in the three-dimensional trend graph of the historical vibration;

[0155] If there is a vibration vector in the three-dimensional trend graph of the historical vibration during the connection vector monitoring period, then there is a vibration connection vector in the three-dimensional trend graph of the historical vibration.

[0156] It can be explained that the current monitoring moment refers to the current moment.

[0157] S6. Extract characteristic monitoring data from the three-dimensional trend graph of the vibration vector, the three-dimensional trend graph of the swing vector, and the three-dimensional trend graph of the pulsation vector in sequence.

[0158] It can be understood that the characteristic monitoring data refers to the coordinate data of the termination points of each vector in the three-dimensional trend graph of the vibration vector, the three-dimensional trend graph of the swing vector, and the three-dimensional trend graph of the pulsation vector.

[0159] ​In the embodiments of the present invention, the extraction of characteristic monitoring data from the three-dimensional trend diagram of the vibration vector, the three-dimensional trend diagram of the swing vector, and the three-dimensional trend diagram of the pulsation vector in sequence includes:

[0160] According to the preset vibration sampling value, swing sampling value, and pulsation sampling value respectively, collect the vibration turning point set, swing turning point set, and pulsation turning point set in the three-dimensional trend diagram of the vibration vector, the three-dimensional trend diagram of the swing vector, and the three-dimensional trend diagram of the pulsation vector;

[0161] Identify the vibration sample point coordinates of each vibration turning point in the collected vibration turning point set to obtain a vibration coordinate set, identify the swing sample point coordinates of each swing turning point in the swing turning point set to obtain a swing coordinate set, and identify the pulsation sample point coordinates of each pulsation turning point in the pulsation turning point set to obtain a pulsation coordinate set;

[0162] Use the vibration coordinate set, swing coordinate set, and pulsation coordinate set as characteristic monitoring data.

[0163] It is understandable that the vibration sampling value, swing sampling value, and pulsation sampling value respectively refer to the sampling point intervals of the vibration turning point, swing turning point, and pulsation turning point. For example: when the vibration sampling value is 8 and the three-dimensional trend diagram of the vibration vector has 2000 points, the 1st, 10th... and 2000th vibration turning points are all in the vibration turning point set. The same applies to the swing sampling value and the pulsation sampling value.

[0164] S7. Input the characteristic monitoring data into a pre-constructed target neural network for fault determination to obtain node determination data.

[0165] It is understandable that the target neural network refers to the original neural network that has been trained.

[0166] In the embodiments of the present invention, before inputting the characteristic monitoring data into a pre-constructed target neural network for fault determination to obtain node determination data, the method further includes:

[0167] Extract the vibration training trend diagram, swing training trend diagram, and pulsation training trend diagram from a pre-constructed standard training trend diagram set in sequence;

[0168] According to the vibration sampling value, swing sampling value, and pulsation sampling value, segmentally extract the vibration training coordinate set, swing training coordinate set, and pulsation training coordinate set from the vibration training trend diagram, swing training trend diagram, and pulsation training trend diagram respectively;

[0169] Train a pre-constructed original neural network using the vibration training coordinate set, the swing training coordinate set, and the pulsation training coordinate set to obtain iterative output data, where the original neural network includes an input layer, a hidden layer, and an output layer. The input layer further includes vibration coordinate input nodes 1, 2, …, z; swing coordinate input nodes 1, 2, …, b; pulsation coordinate input nodes 1, 2, …, m. The output layer further includes a fault type output node, a fault level output node, a first time development level prediction node, a second time development level prediction node, …, an f-th time development level prediction node;

[0170] Obtain the fault type, fault level, first time fault development level, second time fault development level, …, f-th time fault development level of the vibration training trend graph, the swing training trend graph, and the pulsation training trend graph;

[0171] According to the iterative output data, use the fault type, fault level, first time fault development level, second time fault development level, …, f-th time fault development level to iteratively adjust the parameters of the original neural network to obtain a target neural network.

[0172] It can be understood that the fault type output node refers to the node that outputs the characteristic value of the fault type; the fault level output node refers to the node that outputs the characteristic value of the fault level; the first time development level prediction node refers to the node that outputs the development level of the fault at the preset first time; the second time development level prediction node refers to the node that outputs the development level of the fault at the preset second time; the f-th time development level prediction node refers to the node that outputs the development level of the fault at the preset f-th time. The first time can be 1 min, the second time can be 5 min, the third time can be 10 min, and so on.

[0173] It can be understood that without human intervention, a certain type of fault may deteriorate over time. For example, a certain fault may cause the turbine blade to break, and before the turbine blade breaks, a series of fault levels may occur. For example, the turbine blade is damaged after 1 min, partially broken after 5 min, and completely broken after 10 min, etc. Therefore, the first time fault development level, the second time fault development level, …, the f-th time fault development level can be set.

[0174] S8. Judge whether the hydropower unit has a fault according to the node determination data.

[0175] If there is no fault in the hydropower unit, return to the above step of obtaining the real-time monitoring trend atlas of the hydropower unit.

[0176] Understandably, if there is no fault in the hydropower unit, cyclic monitoring is required.

[0177] If there is a fault in the hydropower unit, execute S9, identify the fault type and fault level of the hydropower unit according to the node determination data and conduct an operation trend prediction to obtain a fault development prediction diagram, and complete the fault detection of the hydropower unit based on deep learning.

[0178] Understandably, the fault development prediction diagram refers to a bar chart representing a series of fault levels. For example: at 1 minute, it corresponds to the column of the damaged water turbine blade; at 5 minutes, it corresponds to the column of the partially broken water turbine blade; at 10 minutes, it corresponds to the column of the completely broken water turbine blade. Combine the column of the damaged water turbine blade, the column of the partially broken water turbine blade, and the column of the completely broken water turbine blade into a bar chart to obtain the fault development prediction diagram.

[0179] In the embodiment of the present invention, the step of identifying the fault type and fault level of the hydropower unit according to the node determination data and conducting an operation trend prediction to obtain a fault development prediction diagram includes:

[0180] Successively extract the fault type output value, fault level output value, first-time development level prediction value, second-time development level prediction value,..., f-time development level prediction value of the fault type output node, fault level output node, first-time development level prediction node, second-time development level prediction node,..., f-time development level prediction node from the node determination data;

[0181] Determine the fault type of the hydropower unit according to the fault type output value, determine the fault level of the hydropower unit according to the fault level output value, and determine the first-time fault development level, second-time fault development level,..., f-time fault development level of the hydropower unit according to the first-time development prediction value, second-time development prediction value,..., f-time development prediction value;

[0182] Draw the fault development prediction diagram according to the first-time fault development level, second-time fault development level,..., f-time fault development level.

[0183] Understandably, each output layer node of the target neural network will output a series of numerical values, and then judge the meaning of the numerical value according to the corresponding relationship between the numerical value and the relevant result.

[0184] In an embodiment of the present invention, after determining the fault type and fault level of the hydro-generating unit according to the node determination data and performing an operation trend prediction to obtain a fault development prediction diagram, the method further includes:

[0185] Obtain the current time, and determine the current fault development level of the current time in the fault development prediction diagram;

[0186] Identify the current warning level corresponding to the current fault development level, and perform a fault warning according to the current warning level.

[0187] It is understandable that different current warning levels require different degrees of warning. When the situation is relatively critical, the warning level should be increased to prompt the guardians of the hydro-generating unit to conduct immediate control.

[0188] Compared with the phenomena described in the background art, in the embodiments of the present invention, it is first necessary to obtain a real-time monitoring trend atlas of the hydropower unit, and the real-time monitoring trend atlas includes an upper frame vibration trend chart, a lower frame vibration trend chart, a top cover vibration trend chart, an upper guide swing trend chart, a lower guide swing trend chart, a water guide swing trend chart, a spiral case water pressure pulsation trend chart, a top cover water pressure pulsation trend chart, and a draft tube water pressure pulsation trend chart. Since it is necessary to monitor the hydropower unit in real time, it is necessary to identify the current vibration vector sets of the upper frame vibration trend chart, the lower frame vibration trend chart, and the top cover vibration trend chart. The current vibration vector sets contain real-time change information of the operating state of the hydropower unit. At this time, the target vibration vector can be synthesized in the pre-constructed vibration three-dimensional coordinate system according to the pre-constructed vector synthesis formula and the current vibration vector sets, so as to achieve the goal of structurally processing the current vibration vector. Similarly, it is necessary to identify the current swing vector sets of the upper guide swing trend chart, the lower guide swing trend chart, and the water guide swing trend chart, synthesize the target swing vector in the swing three-dimensional coordinate system according to the current swing vector sets, and identify the current pulsation vector sets of the spiral case water pressure pulsation trend chart, the top cover water pressure pulsation trend chart, and the draft tube water pressure pulsation trend chart, and then synthesize the target pulsation vector in the pulsation three-dimensional coordinate system according to the current pulsation vector sets. Since streamlining the real-time monitoring trend atlas can reduce the complexity of the target neural network during training, the target vibration vector, the target swing vector, and the target pulsation vector can be further streamlined. At this time, the vibration vector three-dimensional trend chart, the swing vector three-dimensional trend chart, and the pulsation vector three-dimensional trend chart can be constructed respectively according to the target vibration vector, the target swing vector, and the target pulsation vector. Finally, the characteristic monitoring data can be extracted from the vibration vector three-dimensional trend chart, the swing vector three-dimensional trend chart, and the pulsation vector three-dimensional trend chart. After obtaining the characteristic monitoring data, the characteristic monitoring data can be input into the pre-constructed target neural network for fault determination to obtain node determination data, and then it can be judged whether the hydropower unit has a fault according to the node determination data. If the hydropower unit has a fault, the fault type and fault level of the hydropower unit can be further identified according to the node determination data and the operation trend can be predicted to obtain a fault development prediction chart, so as to complete the fault detection of the hydropower unit based on deep learning, so as to solve the problems of low fault analysis efficiency and poor early warning accuracy in the current fault detection of hydropower units using deep learning algorithms.

[0189] Embodiment 2:

[0190] As Figure 2 shown, it is a functional module diagram of a fault detection system for hydropower units based on deep learning provided by an embodiment of the present invention.

[0191] The fault detection system 100 for hydropower units based on deep learning according to the present invention can be installed in an electronic device. According to the implemented functions, the fault detection system 100 for hydropower units based on deep learning can include a real-time monitoring trend atlas acquisition module 101, a target vector synthesis module 102, a vector three-dimensional trend graph construction module 103, a feature monitoring data extraction module 104, and a fault determination and prediction module 105. The modules in the present invention can also be referred to as units, which refer to a series of computer program segments that can be executed by a processor of an electronic device and can complete fixed functions, and are stored in the memory of the electronic device.

[0192] The real-time monitoring trend atlas acquisition module 101 is used to acquire the real-time monitoring trend atlas of the hydropower unit, where the real-time monitoring trend atlas includes an upper frame vibration trend graph, a lower frame vibration trend graph, a top cover vibration trend graph, an upper guide swing trend graph, a lower guide swing trend graph, a water guide swing trend graph, a spiral case water pressure pulsation trend graph, a top cover water pressure pulsation trend graph, and a draft tube water pressure pulsation trend graph;

[0193] The target vector synthesis module 102 is used to identify the current vibration vector set of the upper frame vibration trend graph, the lower frame vibration trend graph, and the top cover vibration trend graph, and synthesize a target vibration vector in a pre-constructed vibration three-dimensional coordinate system according to a pre-constructed vector synthesis formula and the current vibration vector set, where the vector synthesis formula is as follows:

[0194]

[0195] Where, represents the target vibration vector, represents the current vibration vector of the upper frame in the current vibration vector set, represents the current vibration vector of the lower frame in the current vibration vector set, represents the current vibration vector of the top cover in the current vibration vector set; identify the current swing vector set of the upper guide swing trend graph, the lower guide swing trend graph, and the water guide swing trend graph, and synthesize a target swing vector in the swing three-dimensional coordinate system according to the current swing vector set; identify the current pulsation vector set of the spiral case water pressure pulsation trend graph, the top cover water pressure pulsation trend graph, and the draft tube water pressure pulsation trend graph, and synthesize a target pulsation vector in the pulsation three-dimensional coordinate system;

[0196] The vector three-dimensional trend graph construction module 103 is used to construct a vibration vector three-dimensional trend graph, a swing vector three-dimensional trend graph, and a pulsation vector three-dimensional trend graph according to the target vibration vector, the target swing vector, and the target pulsation vector respectively;

[0197] The feature monitoring data extraction module 104 is configured to sequentially extract feature monitoring data from the three-dimensional trend graph of the vibration vector, the three-dimensional trend graph of the swing vector, and the three-dimensional trend graph of the pulsation vector;

[0198] The fault determination and prediction module 105 is configured to input the feature monitoring data into a pre-constructed target neural network for fault determination to obtain node determination data; determine whether the hydropower unit has a fault according to the node determination data; if the hydropower unit does not have a fault, return to the above step of obtaining the real-time monitoring trend atlas of the hydropower unit; if the hydropower unit has a fault, identify the fault type and fault level of the hydropower unit according to the node determination data and perform an operation trend prediction to obtain a fault development prediction graph.

[0199] Specifically, each module in the fault detection system 100 of the hydropower unit implemented based on deep learning in the embodiment of the present invention adopts the same technical means as those in the Figure 1 hydropower unit fault detection method implemented based on deep learning described above, and can produce the same technical effects, which will not be elaborated here.

[0200] Embodiment 3:

[0201] As Figure 3 shown, it is a schematic structural diagram of an electronic device for implementing a hydropower unit fault detection method based on deep learning provided by an embodiment of the present invention.

[0202] The electronic device 1 may include a processor 10, a memory 11, a bus 12, and a communication interface 13, and may further include a computer program stored in the memory 11 and executable on the processor 10, such as the program of the hydropower unit fault detection method implemented based on deep learning described in Embodiment 1.

[0203] Among them, the memory 11 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, mobile hard disk, multimedia card, card-type memory (such as SD or DX memory, etc.), magnetic memory, magnetic disk, optical disc, etc. The memory 11 can be an internal storage unit of the electronic device 1 in some embodiments, such as the mobile hard disk of the electronic device 1. The memory 11 can also be an external storage device of the electronic device 1 in some other embodiments, such as a plug-in mobile hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. equipped on the electronic device 1. Further, the memory 11 can also include both the internal storage unit and the external storage device of the electronic device 1. The memory 11 can be used not only to store the application software and various types of data installed in the electronic device 1, such as the code of the fault detection program for hydropower units implemented based on deep learning, etc., but also to temporarily store the data that has been output or will be output.

[0204] The processor 10 can be composed of integrated circuits in some embodiments. For example, it can be composed of a single packaged integrated circuit, or can also be composed of multiple integrated circuits with the same or different functions packaged, including a combination of one or more Central Processing Units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips, etc. The processor 10 is the control core (Control Unit) of the electronic device, connecting all components of the entire electronic device through various interfaces and lines, and by running or executing the programs or modules stored in the memory 11 (such as the program of the fault detection method for hydropower units implemented based on deep learning described in Embodiment 1, etc.), and by calling the data stored in the memory 11, to execute various functions of the electronic device 1 and process data.

[0205] The bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This bus can be divided into an address bus, a data bus, a control bus, etc. The bus is set to realize the connection and communication between the memory 11 and at least one processor 10, etc.

[0206] Figure 3 Only the electronic device with components is shown. Those skilled in the art can understand that Figure 3The structure shown does not constitute a limitation on the electronic device 1, and it may include fewer or more components than those shown, or combine certain components, or have a different component arrangement.

[0207] For example, although not shown, the electronic device 1 may further include devices such as a power supply (such as a battery) for powering each component, a network interface, a user interface, etc. The water turbine generator set fault detection program implemented based on deep learning stored in the memory 11 of the electronic device 1 is a combination of multiple instructions, and when running in the processor 10, it can implement the water turbine generator set fault detection method based on deep learning described in Embodiment 1.

[0208] Specifically, for the specific implementation method of the above instructions by the processor 10, reference can be made to Figures 1 to 2 the description of the relevant steps in the corresponding embodiment, which will not be elaborated here.

[0209] Furthermore, if the modules / units integrated in the electronic device 1 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, the computer-readable medium may include: any entity or system, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory) that can carry the computer program code.

[0210] The present invention also provides a computer-readable storage medium, and the readable storage medium stores a computer program, and when the computer program is executed by a processor of an electronic device, it can implement the water turbine generator set fault detection method based on deep learning described in Embodiment 1.

[0211] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and without departing from the spirit or basic characteristics of the present invention, the present invention can be implemented in other specific forms.

[0212] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A fault detection method for hydropower units implemented based on deep learning, characterized in that, The method includes: Obtaining a real-time monitoring trend atlas of a hydropower unit, where the real-time monitoring trend atlas includes an upper frame vibration trend graph, a lower frame vibration trend graph, a top cover vibration trend graph, an upper guide swing trend graph, a lower guide swing trend graph, a water guide swing trend graph, a spiral case water pressure pulsation trend graph, a top cover water pressure pulsation trend graph, and a draft tube water pressure pulsation trend graph; Identifying a current vibration vector set of the upper frame vibration trend graph, the lower frame vibration trend graph, and the top cover vibration trend graph, and synthesizing a target vibration vector in a pre-constructed vibration three-dimensional coordinate system according to a pre-constructed vector synthesis formula and the current vibration vector set, where the vector synthesis formula is as follows: Among them, represents the target vibration vector, represents the current vibration vector of the upper machine rack in the current vibration vector set, represents the current vibration vector of the lower machine rack in the current vibration vector set, represents the current vibration vector of the top cover in the current vibration vector set; Identifying a current swing vector set of the upper guide swing trend graph, the lower guide swing trend graph, and the water guide swing trend graph, and synthesizing a target swing vector in the swing three-dimensional coordinate system according to the current swing vector set; Identifying a current pulsation vector set of the spiral case water pressure pulsation trend graph, the top cover water pressure pulsation trend graph, and the draft tube water pressure pulsation trend graph, and synthesizing a target pulsation vector in the pulsation three-dimensional coordinate system according to the current pulsation vector set; Constructing a vibration vector three-dimensional trend graph, a swing vector three-dimensional trend graph, and a pulsation vector three-dimensional trend graph respectively according to the target vibration vector, the target swing vector, and the target pulsation vector; Successively extracting characteristic monitoring data from the vibration vector three-dimensional trend graph, the swing vector three-dimensional trend graph, and the pulsation vector three-dimensional trend graph; Inputting the characteristic monitoring data into a pre-constructed target neural network for fault determination to obtain node determination data; Judging whether the hydropower unit has a fault according to the node determination data; If the hydropower unit has no fault, return to the step of obtaining the real-time monitoring trend atlas of the hydropower unit above; If the hydropower unit has a fault, identify the fault type and fault level of the hydropower unit according to the node determination data and perform an operation trend prediction to obtain a fault development prediction graph, completing the fault detection of the hydropower unit based on deep learning.

2. The fault detection method for a hydropower unit according to claim 1, characterized in that The identification of the current vibration vector set of the upper frame vibration trend graph, the lower frame vibration trend graph, and the top cover vibration trend graph includes: Identifying the current monitoring signal points of the upper frame of the upper frame vibration trend graph; According to a preset vibration sampling period and the current monitoring signal points of the upper frame, extracting adjacent monitoring signal points of the upper frame in the upper frame vibration trend graph; Constructing a current vibration vector of the upper frame with the adjacent monitoring signal points of the upper frame as the starting point and the current monitoring signal points of the upper frame as the ending point; Identifying the current monitoring signal points of the lower frame of the lower frame vibration trend graph; According to the vibration sampling period and the current monitoring signal points of the lower frame, extracting adjacent monitoring signal points of the lower frame in the lower frame vibration trend graph; Constructing a current vibration vector of the lower frame with the adjacent monitoring signal points of the lower frame as the starting point and the current monitoring signal points of the lower frame as the ending point; Identifying the current monitoring signal points of the top cover of the top cover vibration trend graph; According to the vibration sampling period and the current monitoring signal points of the top cover, extracting adjacent monitoring signal points of the top cover in the top cover vibration trend graph; Taking the monitoring signal point adjacent to the top cover as the starting point and the current monitoring signal point of the top cover as the ending point, construct the current vibration vector of the top cover; Summarize the current vibration vectors of the upper frame, the current vibration vectors of the lower frame and the current vibration vector of the top cover to obtain the current vibration vector set.

3. The fault detection method for a hydropower unit according to claim 2, characterized in that The synthesizing of the target vibration vector in the pre-constructed three-dimensional vibration coordinate system according to the pre-constructed vector synthesis formula and the current vibration vector set includes: Extract the current vibration vector of the upper frame from the current vibration vector set and identify the upper frame coordinate plane of the three-dimensional vibration coordinate system; Fill the current vibration vector of the upper frame into the upper frame coordinate plane to obtain the vector to be synthesized for the upper frame, and the starting point of the vector to be synthesized for the upper frame coincides with the origin of the three-dimensional vibration coordinate system; Extract the current vibration vector of the lower frame from the current vibration vector set and identify the lower frame coordinate plane of the three-dimensional vibration coordinate system; Fill the current vibration vector of the lower frame into the lower frame coordinate plane to obtain the vector to be synthesized for the lower frame, and the starting point of the vector to be synthesized for the lower frame coincides with the origin of the three-dimensional vibration coordinate system; Extract the current vibration vector of the top cover from the current vibration vector set and identify the top cover coordinate plane of the three-dimensional vibration coordinate system; Fill the current vibration vector of the top cover into the top cover coordinate plane to obtain the vector to be synthesized for the top cover, and the starting point of the vector to be synthesized for the top cover coincides with the origin of the three-dimensional vibration coordinate system; According to the vector to be synthesized for the upper frame, the vector to be synthesized for the lower frame and the vector to be synthesized for the top cover, use the vector synthesis formula to synthesize the target vibration vector in the three-dimensional vibration coordinate system.

4. The method for detecting faults of a hydropower unit according to claim 1, characterized in that, The constructing of the three-dimensional trend diagram of the vibration vector, the three-dimensional trend diagram of the deflection vector and the three-dimensional trend diagram of the pulsation vector respectively according to the target vibration vector, the target deflection vector and the target pulsation vector includes: Obtain the historical three-dimensional trend diagram of the vibration, and judge whether there is a preset vibration connection vector in the historical three-dimensional trend diagram of the vibration, where the vibration connection vector refers to the vibration vector in the historical three-dimensional trend diagram of the vibration that can be connected to the target vibration vector; If there is no vibration connection vector in the historical three-dimensional trend diagram of the vibration, translate the starting point of the target vibration vector to the coordinate origin of the historical three-dimensional trend diagram of the vibration to obtain the three-dimensional trend diagram of the vibration vector; If there is a vibration connection vector in the historical three-dimensional trend diagram of the vibration, identify the ending point of the vibration connection vector; Connect the target vibration vector to the ending point of the vibration connection vector to obtain the three-dimensional trend diagram of the vibration vector; Obtain the historical three-dimensional trend diagram of the deflection, and judge whether there is a preset deflection connection vector in the historical three-dimensional trend diagram of the deflection, where the deflection connection vector refers to the deflection vector in the historical three-dimensional trend diagram of the deflection that can be connected to the target deflection vector; If there is no deflection connection vector in the historical three-dimensional trend diagram of the deflection, translate the starting point of the target deflection vector to the coordinate origin of the historical three-dimensional trend diagram of the deflection to obtain the three-dimensional trend diagram of the deflection vector; If there is a deflection connection vector in the historical three-dimensional trend diagram of the deflection, identify the ending point of the deflection connection vector; Connect the target deflection vector to the termination point of the deflection connection vector to obtain a three-dimensional trend graph of the deflection vector; Obtain a historical pulsation three-dimensional trend graph, and determine whether there is a preset pulsation connection vector in the historical pulsation three-dimensional trend graph, where the pulsation connection vector refers to a pulsation vector in the historical vibration three-dimensional trend graph that can be connected to the target pulsation vector; If there is no pulsation connection vector in the historical pulsation three-dimensional trend graph, translate the starting point of the target pulsation vector to the coordinate origin of the historical pulsation three-dimensional trend graph to obtain a three-dimensional trend graph of the pulsation vector; If there is a pulsation connection vector in the historical pulsation three-dimensional trend graph, identify the termination point of the pulsation connection vector; Connect the target pulsation vector to the termination point of the pulsation connection vector to obtain a three-dimensional trend graph of the pulsation vector.

5. The fault detection method for a hydropower unit according to any one of claims 1-4, characterized in that, The determination of whether there is a preset vibration connection vector in the historical vibration three-dimensional trend graph includes: Obtain the current monitoring time, and construct a connection vector monitoring period according to the vibration sampling period using the following formula: X j = [t n - T, t n ​ Among them, X j represents the connection vector monitoring period, t n represents the current monitoring moment, and T represents the vibration sampling period; Determine whether there is a vibration vector in the historical vibration three-dimensional trend graph during the connection vector monitoring period; If there is no vibration vector in the historical vibration three-dimensional trend graph during the connection vector monitoring period, then there is no vibration connection vector in the historical vibration three-dimensional trend graph; If there is a vibration vector in the historical vibration three-dimensional trend graph during the connection vector monitoring period, then there is a vibration connection vector in the historical vibration three-dimensional trend graph.

6. The method for detecting faults of a hydropower unit according to claim 4, characterized in that, The sequential extraction of characteristic monitoring data from the vibration vector three-dimensional trend graph, the deflection vector three-dimensional trend graph, and the pulsation vector three-dimensional trend graph includes: Respectively collect a vibration turning point set, a deflection turning point set, and a pulsation turning point set in the vibration vector three-dimensional trend graph, the deflection vector three-dimensional trend graph, and the pulsation vector three-dimensional trend graph according to preset vibration sampling values, deflection sampling values, and pulsation sampling values; Identify the vibration sample point coordinates of each collected vibration turning point in the collected vibration turning point set to obtain a vibration coordinate set, identify the deflection sample point coordinates of each deflection turning point in the deflection turning point set to obtain a deflection coordinate set, and identify the pulsation sample point coordinates of each pulsation turning point in the pulsation turning point set to obtain a pulsation coordinate set; Use the vibration coordinate set, the deflection coordinate set, and the pulsation coordinate set as characteristic monitoring data.

7. The fault detection method for a hydropower unit according to claim 6, characterized in that Before inputting the characteristic monitoring data into a pre-constructed target neural network for fault determination to obtain node determination data, the method further includes: Sequentially extract a vibration training trend graph, a deflection training trend graph, and a pulsation training trend graph from a pre-constructed standard training trend graph set; Segmentally extract a vibration training coordinate set, a deflection training coordinate set, and a pulsation training coordinate set from the vibration training trend graph, the deflection training trend graph, and the pulsation training trend graph respectively according to the vibration sampling value, the deflection sampling value, and the pulsation sampling value; Train a pre - constructed original neural network using the vibration training coordinate set, the swing training coordinate set, and the pulsation training coordinate set to obtain iterative output data. The original neural network includes an input layer, a hidden layer, and an output layer. The input layer further includes vibration coordinate input nodes 1, vibration coordinate input nodes 2, …, vibration coordinate input nodes z, swing coordinate input nodes 1, swing coordinate input nodes 2, …, swing coordinate input nodes b, pulsation coordinate input nodes 1, pulsation coordinate input nodes 2, …, pulsation coordinate input nodes m. The output layer further includes a fault type output node, a fault level output node, a first - time development level prediction node, a second - time development level prediction node, …, an f - th time development level prediction node; Obtain the fault type, fault level, first - time fault development level, second - time fault development level, …, f - th time fault development level of the vibration training trend graph, the swing training trend graph, and the pulsation training trend graph; According to the iterative output data, use the fault type, fault level, first - time fault development level, second - time fault development level, …, f - th time fault development level to perform iterative parameter adjustment on the original neural network to obtain a target neural network.

8. The method for detecting faults of a hydroelectric generating unit according to claim 1, characterized in that, The method of identifying the fault type and fault level of the hydropower unit based on the node determination data and performing an operation trend prediction to obtain a fault development prediction graph includes: Successively extract the fault type output value, fault level output value, first - time development prediction value, second - time development prediction value, …, f - th time development prediction value of the fault type output node, fault level output node, first - time development level prediction node, second - time development level prediction node, …, f - th time development level prediction node from the node determination data; Determine the fault type of the hydropower unit according to the fault type output value, determine the fault level of the hydropower unit according to the fault level output value, and determine the first - time fault development level, second - time fault development level, …, f - th time fault development level of the hydropower unit according to the first - time development prediction value, second - time development prediction value, …, f - th time development prediction value; Draw the fault development prediction graph according to the first - time fault development level, second - time fault development level, …, f - th time fault development level.

9. The method for detecting faults of a hydropower unit according to claim 8, characterized in that, After the method of identifying the fault type and fault level of the hydropower unit based on the node determination data and performing an operation trend prediction to obtain a fault development prediction graph, the method further includes: Obtain the current time and judge the current fault development level of the current time in the fault development prediction graph; Identify the current warning level corresponding to the current fault development level and perform a fault warning according to the current warning level.

10. A hydroelectric unit fault detection system implemented based on deep learning, characterized in that, The system includes: A real-time monitoring trend atlas acquisition module for acquiring a real-time monitoring trend atlas of a hydropower unit, where the real-time monitoring trend atlas includes an upper frame vibration trend chart, a lower frame vibration trend chart, a top cover vibration trend chart, an upper guide swing trend chart, a lower guide swing trend chart, a water guide swing trend chart, a spiral case water pressure pulsation trend chart, a top cover water pressure pulsation trend chart, and a draft tube water pressure pulsation trend chart; A target vector synthesis module for identifying a current vibration vector set of the upper frame vibration trend chart, the lower frame vibration trend chart, and the top cover vibration trend chart, and synthesizing a target vibration vector in a pre-constructed vibration three-dimensional coordinate system according to a pre-constructed vector synthesis formula and the current vibration vector set, where the vector synthesis formula is as follows: Among them, represents the target vibration vector, represents the current vibration vector of the upper frame in the current vibration vector set, represents the current vibration vector of the lower frame in the current vibration vector set, represents the current vibration vector of the top cover in the current vibration vector set; identify the current set of deflection vectors of the upper guide deflection trend graph, lower guide deflection trend graph, and water guide deflection trend graph, and synthesize the target deflection vector in the deflection three-dimensional coordinate system according to the current set of deflection vectors; identify the current set of pulsation vectors of the spiral case water pressure pulsation trend graph, top cover water pressure pulsation trend graph, and draft tube water pressure pulsation trend graph, and synthesize the target pulsation vector in the pulsation three-dimensional coordinate system according to the current set of pulsation vectors; A vector three-dimensional trend chart construction module for constructing a vibration vector three-dimensional trend chart, a swing vector three-dimensional trend chart, and a pulsation vector three-dimensional trend chart respectively according to the target vibration vector, the target swing vector, and the target pulsation vector; A feature monitoring data extraction module for sequentially extracting feature monitoring data from the vibration vector three-dimensional trend chart, the swing vector three-dimensional trend chart, and the pulsation vector three-dimensional trend chart; A fault determination and prediction module for inputting the feature monitoring data into a pre-constructed target neural network for fault determination to obtain node determination data; judging whether the hydropower unit has a fault according to the node determination data; if the hydropower unit has no fault, returning to the step of acquiring the real-time monitoring trend atlas of the hydropower unit as described above; if the hydropower unit has a fault, identifying the fault type and fault level of the hydropower unit according to the node determination data and performing an operation trend prediction to obtain a fault development prediction chart.

Images