A method and device for evaluating the winding state of a synchronous generator
Through deep neural network combining vibration, temperature and electrical data for synchronous generator winding status evaluation, the problem of traditional methods relying on offline testing is solved, and non-invasive, accurate and efficient winding status evaluation is achieved, reducing maintenance costs and improving maintenance Real-time performance of evaluation.
Patent Information
- Application Number
- CN202411022601.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-29
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2044-07-29
AI Technical Summary
The traditional synchronous generator winding state evaluation method relies on offline testing, requires generator downtime, affects power generation efficiency and increases maintenance costs, and lacks non-invasive real-time evaluation capabilities.
Deep neural network is used to combine vibration, temperature and electrical data for winding state evaluation, collect data through non-invasive sensors, build a multi-layer feedforward neural network for online evaluation, and improve evaluation accuracy using discrete Fourier transform and data preprocessing. , possess dynamic learning and self-optimization capabilities.
It realizes accurate, scientific and efficient evaluation of the winding status of the synchronous generator, reduces maintenance costs, improves the accuracy and real-time evaluation, and has the ability to learn online.
Smart Images

Figure CN118551313B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of synchronous generators, and in particular to a method and a device for evaluating the winding state of a synchronous generator. Background Art
[0002] The generator rotor winding of the synchronous generator uses an AC exciter to provide excitation current, which has the advantages of compact structure, simple circuit, reliable operation, convenient control and low maintenance. This structure eliminates brushes and slip rings, does not generate commutation sparks, and is suitable for operation in dangerous working environments. At the same time, it also improves the insulation life of the motor, eliminates maintenance work and improves the reliability of the excitation system. The brushless excitation system can be used in special occasions with explosion-proof, dust-proof, and corrosion-proof requirements, and is widely used in mining, petroleum, chemical and other fields. The brushless excitation system has high stability and is one of the common excitation methods for large-capacity generator sets.
[0003] Accurate assessment of the winding condition of synchronous generators is crucial to ensure safe and reliable operation of the power grid. Traditional winding condition assessment methods usually rely on offline tests, such as resistance measurement of different windings, impedance measurement to ground, etc., but these measurements require the generator to be shut down, which affects power generation efficiency and increases maintenance costs. Summary of the invention
[0004] On one hand, the present invention provides a method for evaluating the winding state of a synchronous generator, which determines the winding state in a non-invasive manner, comprising the following steps:
[0005] Step S1: Collect relevant data through synchronous generator sensors;
[0006] Step S2: constructing a deep neural network for synchronous generator winding state assessment;
[0007] Step S3: Use a deep neural network to evaluate the winding state of the synchronous generator.
[0008] Preferably, step S1 specifically includes:
[0009] Step S101: collecting synchronous generator vibration data;
[0010] The vibration data of the synchronous generator is recorded as
[0011] ,
[0012] Where K is the number of vibration sensors, T is the current acquisition time, and N is the number of data points of synchronous generator vibration collected continuously;
[0013] Step S102: collecting synchronous generator temperature data;
[0014] The temperature data of the synchronous generator is recorded as:
[0015] ,
[0016] Where T is the current acquisition time, L is the number of temperature sensors, and Q is the number of data points collected;
[0017] Step S103: collecting electrical data of the synchronous generator;
[0018] The electrical data of the synchronous generator mainly includes the voltage and current data of the stator and rotor, which are recorded as
[0019] , ,
[0020] , ,
[0021] In the formula, DY represents voltage, DL represents current, T represents the acquisition time, subscript D represents the stator end, Z represents the rotor end; U, V, W; u, v, w mark different items of current corresponding to the three items of the generator, which are different winding ends of the generator.
[0022] Preferably, step S2 specifically includes:
[0023] Step S201: Establish a deep neural network structure;
[0024] The input of the neural network structure is the synchronous generator vibration data, the synchronous generator temperature data and the synchronous generator electrical data, and the output is the evaluation value of the winding state;
[0025] Step S202: preprocessing of input data.
[0026] Preferably, step S201 specifically includes:
[0027] Assume that the output value of the evaluation value of the winding state is [E D 、E Z ], that is, the stator winding state evaluation value E D and rotor winding state evaluation value E Z , issue warning information for faults;
[0028] The fault is one or more of turn-to-turn short circuit, three-phase imbalance, and short circuit to ground;
[0029] The collected data of three consecutive time periods T-2, T-1 and T are used as input data.
[0030] The evaluation results at time T-1 and the manually verified information are input into the neural network, and the deep neural network is dynamically adjusted to enable it to have dynamic learning capabilities.
[0031] Preferably, step S202 specifically includes:
[0032] The synchronous generator vibration data is processed by: normalizing N data points acquired by the vibration sensor;
[0033] Take the jth vibration sensor collecting data at time T as an example, , j=1,2,..., K, preprocess it;
[0034] For the jth sensor acquisition area, the maximum allowable value of the vibration amplitude is set as If the vibration amplitude exceeds the maximum allowable value, an alarm will be triggered directly;
[0035] For the above Perform the following processing
[0036] ,in , i=1,2,…,N, j=1,2,…,K;
[0037] And use discrete Fourier transform to obtain its frequency domain parameters and input them into the winding state evaluation network;
[0038] ,
[0039] Among them, N is the number of points of the collected vibration signal, n is the number of the discrete signal in the time domain, m is the number of the frequency domain signal, and the number of points of the frequency domain signal is also N; i is an imaginary number representing the symbol, ;
[0040] All the data from different sensors are processed and input into the winding state evaluation network. The vibration data of all processed input networks is recorded as Data_ZD.
[0041] Preferably, step S202 specifically includes:
[0042] The synchronous generator temperature data is processed in the following manner:
[0043] Take the data collected by the jth temperature sensor at time T as an example, , j=1,2,..,Q, preprocess it;
[0044] For the jth sensor collection area, the maximum allowable temperature value is set as , if it exceeds this value, an alarm will be issued directly;
[0045] For the above Perform the following processing
[0046] ,
[0047] in, ;
[0048] After processing, all data are input into the evaluation network; the temperature data of all processed input networks is recorded as Data_WD.
[0049] Preferably, step S202 specifically includes:
[0050] The electrical data of the synchronous generator is processed in the following manner: quantization is performed within a set range, and if the voltage, current, etc. exceed the maximum allowable value, an alarm message is directly issued;
[0051] The normalization method is as follows:
[0052] ,
[0053] ,
[0054] ,
[0055] ,
[0056] In the formula, the characters after max_ represent the corresponding items of the generator, stator and rotor. Indicates the corresponding maximum allowable voltage value, Indicates the corresponding maximum current allowed value;
[0057] After processing according to the above method, all electrical related data are input into the evaluation network; the electrical data of all processed input networks is recorded as Data_DQ.
[0058] Preferably, step S3 specifically includes:
[0059] Train based on historical data and use the trained network to evaluate winding status;
[0060] The data of the input sensors at time T-2, T-1, and T, as well as the evaluation result and manual review result at time T-1, the network output value is the synchronous generator winding state evaluation value [E D 、E Z ] and warning information [a1, a2, a3, a4], where a1-a4 are Boolean numbers 0 or 1;
[0061] A network dynamic learning method based on manual judgment information uses the results of manual verification of fault information as the input of manual judgment information;
[0062] Suppose the abstract mathematical expression of the network is
[0063]
[0064] In the formula, For the calculation of input data, are the weights and biases of the multi-layer network, Multi-layer network calculation results;
[0065] If the output warning information is inconsistent with the actual manual verification, it means that the network output result is not accurate enough. Optimize, the optimization method is as follows:
[0066]
[0067] In the formula, , is the optimized value, , is the learning rate;
[0068] The next time you calculate, use the new , Calculated as network parameters.
[0069] Another aspect of the present invention provides a synchronous generator winding state evaluation device, comprising:
[0070] A data collection module collects relevant data through synchronous generator sensors;
[0071] A data processing module processes the data by constructing a deep neural network for synchronous generator winding state assessment;
[0072] Evaluation module, which evaluates the winding status of synchronous generator.
[0073] Another aspect of the present invention provides a computer storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the above-mentioned method for evaluating the winding state of a synchronous generator are implemented.
[0074] The beneficial effects of the present invention are:
[0075] 1) The present invention adopts a non-invasive method to evaluate the winding state of a synchronous generator, which can adapt to the evaluation method of the winding state of a synchronous generator and greatly improve the accuracy of the evaluation;
[0076] 2) The present invention fully combines the winding state of the synchronous generator and adopts an improved deep neural network method to improve the scientificity and rationality of the evaluation results;
[0077] 3) The evaluation process of the present invention scientifically processes the vibration temperature and electrical parameters of the synchronous generator winding, so that the data input into the deep neural network has a unified quantification method, and adopts different processing methods for the three types of data, which greatly improves the accuracy of the evaluation;
[0078] 4) The evaluation system of the present invention has online self-learning capability. BRIEF DESCRIPTION OF THE DRAWINGS
[0079] Figure 1 Schematic diagram of a synchronous generator winding state assessment method according to Embodiment 1;
[0080] Figure 2 This is a schematic diagram of the method of step S2 of Example 1 of this scheme. DETAILED DESCRIPTION
[0081] 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.
[0082] The embodiment of the present application provides an excitation control system for a brushless excitation synchronous generator based on an optimization algorithm. The present invention provides an online synchronous generator winding state assessment method, which aims to overcome the shortcomings of traditional methods. The method first processes the collected generator characteristic data, and then uses a deep neural network algorithm to analyze and evaluate the processed data, judges the winding state in a non-invasive manner, and indicates possible faults.
[0083] Reference Figure 1 The above is a schematic diagram of a synchronous generator winding state assessment method provided by an embodiment of the present invention.
[0084] Step S1: Collect relevant data through synchronous generator sensors.
[0085] First, various data on the working status of the synchronous generator are obtained through corresponding sensors. Mainly including:
[0086] Step S101: collecting vibration data of the synchronous generator.
[0087] The vibration data is recorded as
[0088] ,
[0089] Among them, K is the number of vibration sensors. Generally, large synchronous generators will have multiple data acquisition sensors. T is the current acquisition time. Since it is online monitoring, it is generally collected once every certain period of time. N is the number of data points of synchronous generator vibration collected continuously.
[0090] Step S102: Collecting synchronous generator temperature data.
[0091] The temperature data of the synchronous generator is recorded as:
[0092] ,
[0093] Where T is the current acquisition time, L is the number of temperature sensors, and Q is the number of collected data points.
[0094] Step S103: Collecting electrical data of the synchronous generator.
[0095] Mainly the voltage and current data of the stator and rotor. The data is recorded as
[0096] , ,
[0097] , ,
[0098] In the formula, DY represents voltage, DL represents current, T represents the acquisition time, subscript D represents the stator end, and Z represents the rotor end. U, V, and W (u, v, and w represent different items of current) correspond to the three items of the generator, which are different winding ends of the generator.
[0099] Step S2: Construct a deep neural network for synchronous generator winding state assessment.
[0100] Step S201: Establish a deep neural network structure.
[0101] The network structure adopts a multi-layer feedforward neural network. The input is the synchronous generator vibration, temperature and electrical data collected by the sensor, and the output is the evaluation value of the winding state. The output value [E D 、E Z ], that is, the stator winding state evaluation value E D and rotor winding state evaluation value E Z , the range is 0-1. The closer to 1, the better the generator winding state is, and the closer to 0, the worse the generator winding state is. For possible faults, such as turn-to-turn short circuit, three-phase imbalance, short circuit to ground, etc., alarm information is issued, represented by 4-bit binary numbers, which can represent 15 types of faults. Currently, the first 1-7 types are mainly used, and subsequent support expansion is available. [0,0,0,0] means no warning information is issued. The corresponding fault output code and meaning are defined in Table 1:
[0102] Table 1 Fault output codes and meaning definitions
[0103]
[0104] The collected data of three consecutive time periods, T-2, T-1, and T, are used as input data, taking into account the historical data of the current moment T and the previous two moments, so that a full comparative analysis can be performed in the deep neural network to improve the scientific nature of the evaluation results. In addition, the evaluation results of the previous moment (T-1) and the information of manual verification are input into the neural network, and the deep neural network is dynamically adjusted to enable it to have the ability of dynamic learning. The verification result H is 0 or 1, where the value of H is 1 if the manual verification is consistent with the actual situation, otherwise it is 0.
[0105] Step S202: preprocessing of input data.
[0106] Figure 2 The vibration data processing network, temperature data processing network and electrical data processing network in the embodiment mainly utilize the known three-phase synchronous generator related parameter setting requirements to perform data preprocessing.
[0107] For example, the allowable values of the generator vibration amplitude and frequency, the allowable temperature values of different parts of the generator, and the rated and maximum values of electrical data such as voltage and current. By preprocessing the above data and then normalizing the data of different dimensions and description forms, the basic input is provided for the deep neural network for winding state assessment.
[0108] (1) Vibration data processing
[0109] Standardize the N data points acquired by the vibration sensor. Take the data collected by the jth vibration sensor at time T as an example, , j=1, 2,..., K, and preprocess it. For the jth sensor acquisition area, the maximum allowable value of the vibration amplitude is set as If the value is exceeded, an alarm will be triggered. Perform the following processing
[0110] ,in , i=1,2,…,N, j=1,2,…,K;
[0111] Discrete Fourier transform is used to obtain some frequency domain parameters, such as fundamental frequency and harmonic components, and input them into the winding state evaluation network.
[0112]
[0113] Among them, N is the number of points of the collected vibration signal (discrete points in the time domain), n is the number of the discrete signal in the time domain (the value range is 1~N), m is the number of the frequency domain signal (the value range is 0~N-1), and the number of points of the frequency domain signal is also N. i is an imaginary number representing the symbol, .
[0114] It should be noted that all data from different sensors need to be processed and then input into the winding state evaluation network. The vibration data of all processed input networks is recorded as Data_ZD.
[0115] The above data processing method greatly improves the accuracy of data entering neural network training and improves the accuracy of evaluation results.
[0116] (2) Temperature processing method
[0117] The processing of temperature data is similar to that of vibration signals, except that temperature does not need to be discrete Fourier transformed because temperature characteristics have no frequency domain attributes. Take the data collected by the jth temperature sensor at time T as an example, , j=1,2,..,Q, and preprocess it. For the jth sensor collection area, the maximum allowable temperature value is recorded as If it exceeds this value, an alarm will be triggered directly.
[0118] For the above Perform the following processing
[0119] ,
[0120] in, .
[0121] After processing, all data are input into the evaluation network. The temperature data of all processed input networks is recorded as Data_WD.
[0122] The above data processing method greatly improves the accuracy of data entering neural network training and improves the accuracy of evaluation results.
[0123] (3) Processing of electrical data
[0124] The electrical data processing method is similar to the previous two processing ideas. It is quantified within the set range. If the voltage, current, etc. exceed the maximum allowable value, an alarm message will be directly issued. The normalization method is as follows:
[0125] ,
[0126] ,
[0127] ,
[0128] ,
[0129] In the formula, the characters after max_ represent the corresponding terms of the generator, stator and rotor, for example etc. represent the corresponding maximum allowable voltage values. Indicates the corresponding maximum current allowed value.
[0130] After processing according to the above method, all electrical related data are input into the evaluation network. The electrical data of all processed input networks is recorded as Data_DQ.
[0131] The above data processing method greatly improves the accuracy of data entering neural network training and improves the accuracy of evaluation results.
[0132] Step S3: Use a deep neural network to evaluate the winding state of the synchronous generator.
[0133] The training is carried out according to the historical data, and then the trained network is used to evaluate the winding state. The data of sensors T-2, T-1, and T, as well as the evaluation results and manual review results at T-1, are input, and the network output value is the evaluation value of the synchronous generator winding state [E D 、E Z ] and warning information [a1,a2,a3,a4], where a1-a4 are Boolean numbers of 0 or 1.
[0134] Preferably, the present invention proposes a network dynamic learning method based on manual judgment information, using the results of manual verification of fault information as the input of manual judgment information. Suppose the abstract mathematical expression of the network is
[0135]
[0136] In the formula, For the calculation of input data, are the weights and biases of the multi-layer network, Multi-layer network calculation results.
[0137] If the output warning information is inconsistent with the actual manual verification, it means that the network output result is not accurate enough. Optimize, the optimization method is as follows:
[0138]
[0139] In the formula, , is the optimized value, , is the learning rate.
[0140] The next time you calculate, use the new , Calculated as network parameters.
[0141] Preferably, the synchronous generator winding state assessment device provided by the present invention comprises:
[0142] The data collection module collects relevant data through synchronous generator sensors.
[0143] The data processing module processes the data by constructing a deep neural network for synchronous generator winding status evaluation.
[0144] Evaluation module, which evaluates the winding status of synchronous generator.
[0145] Preferably, in one embodiment of the present invention, a computer storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the steps of data fusion in the aforementioned embodiment are implemented.
[0146] Computer storage media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, machine-readable media does not include temporary computer-readable media (transitory media), such as modulated data signals and carrier waves.
[0147] For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.
[0148] In the description of the present invention, unless otherwise specified, the terms "upper", "lower", "left", "right", "inside", "outside", etc. indicate directions or positional relationships based on the directions or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, they cannot be understood as limitations on the present invention.
[0149] Finally, it should be explained that the above technical solution is only one implementation method of the present invention. For those skilled in the art, it is easy to make various types of improvements or modifications based on the application methods and principles disclosed in the present invention, and it is not limited to the method described in the above specific implementation method of the present invention. Therefore, the method described above is only preferred and does not have a restrictive meaning.
Claims
1. A method for evaluating the winding condition of a synchronous generator, Features Determine the winding status in a non-invasive manner, including the following steps: Step S1: Collect relevant data through synchronous generator sensors; Step S2: constructing a deep neural network for synchronous generator winding state assessment; The step S2 specifically includes: Step S201: Establish a deep neural network structure; The input of the neural network structure is the synchronous generator vibration data, the synchronous generator temperature data and the synchronous generator electrical data, and the output is the evaluation value of the winding state; Step S202: preprocessing input data; Step S3: Evaluate the winding state of the synchronous generator by constructing a deep neural network; The step S3 specifically includes: Train based on historical data and use the trained network to evaluate winding status; The data of sensors at time T-2, T-1, and T, as well as the evaluation result and manual review result at time T-1, are input. The network output value is the synchronous generator winding state evaluation value [E D 、E Z ] and warning information [a1, a2, a3, a4], where a1-a4 are Boolean numbers 0 or 1; A network dynamic learning method based on manual judgment information uses the results of manual verification of fault information as the input of manual judgment information; Suppose the abstract mathematical expression of the network is In the formula, is the calculation of input data, W,b is the weight and bias of the multi-layer network, and f(W,b) is the calculation result of the multi-layer network; If the output warning information is inconsistent with the actual manual verification, it means that the network output result is not accurate enough, so W,b is optimized. The optimization method is as follows: Where W * 、b * is the optimized value, α and β are the learning rates; Then the next time you calculate, use the new W * 、b * Calculated as network parameters; The step S202 specifically includes: The synchronous generator vibration data is processed by: normalizing N data points acquired by the vibration sensor; Take the jth vibration sensor collecting data at time T as an example, j=1,2,...,K, preprocess it; For the jth sensor acquisition area, the maximum allowable value of the vibration amplitude is set as max_ZD j If the vibration amplitude exceeds the maximum allowable value, an alarm will be triggered directly; For the above Perform the following processing in i=1,2,…,N,j=1,2,…,K; And use discrete Fourier transform to obtain its frequency domain parameters and input them into the winding state evaluation network; Among them, N is the number of points of the collected vibration signal, n is the number of the discrete signal in the time domain, m is the number of the frequency domain signal, and the number of points of the frequency domain signal is also N; i is the imaginary number representing the sign, i 2 = -1; All the data from different sensors are processed and input into the winding state evaluation network. The vibration data of all processed input networks is recorded as Data_ZD. The synchronous generator temperature data is processed in the following manner: Take the data collected by the jth temperature sensor at time T as an example, j=1,2,..,Q, preprocess it; For the jth sensor collection area, the maximum allowable temperature value is set as max_WD j , if it exceeds this value, an alarm will be issued directly; For the above Perform the following processing in, After processing, all data are input into the evaluation network; the temperature data of all processed input networks is recorded as Data_WD; The electrical data of the synchronous generator is processed in the following manner: quantization is performed within a set range, and if the voltage, current, etc. exceed the maximum allowable value, an alarm message is directly issued; The normalization method is as follows: In the formula, the characters after max_ represent the corresponding items of the generator, stator and rotor, max_U D Indicates the corresponding maximum allowable voltage value, max_Iu D Indicates the corresponding maximum current allowed value; After processing according to the above method, all electrical related data are input into the evaluation network; the electrical data of all processed input networks is recorded as Data_DQ.
2. The method according to claim 1, It is characterized in that The step S1 specifically includes: Step S101: collecting vibration data of synchronous generator; The vibration data of the synchronous generator is recorded as Where K is the number of vibration sensors, T is the current acquisition time, and N is the number of data points of synchronous generator vibration collected continuously; Step S102: collecting synchronous generator temperature data; The temperature data of the synchronous generator is recorded as: Where T is the current acquisition time, L is the number of temperature sensors, and Q is the number of data points collected; Step S103: collecting electrical data of the synchronous generator; The electrical data of the synchronous generator mainly includes the voltage and current data of the stator and rotor, which are recorded as In the formula, DY represents voltage, DL represents current, T represents the acquisition time, subscript D represents the stator end, Z represents the rotor end; U, V, W; u, v, w mark different items of current corresponding to the three items of the generator, which are different winding ends of the generator.
3. The method according to claim 2, It is characterized in that The step S201 specifically includes: Assume that the output value of the evaluation value of the winding state is [E D 、E Z ], that is, the stator winding state evaluation value E D and rotor winding state evaluation value E Z , issue warning information for faults; The fault is one or more of turn-to-turn short circuit, three-phase imbalance, and short circuit to ground; The collected data of three consecutive time periods T-2, T-1 and T are used as input data. The evaluation results at time T-1 and the manually verified information are input into the neural network, and the deep neural network is dynamically adjusted to enable it to have dynamic learning capabilities.
4. A synchronous generator winding state assessment device, used to perform the method according to any one of claims 1 to 3, Features include: A data collection module collects relevant data through synchronous generator sensors; A data processing module processes the data by constructing a deep neural network for synchronous generator winding state assessment; Evaluation module, which evaluates the winding status of synchronous generator.
5. A computer storage medium having a computer program stored thereon, Features: When the computer program is executed by a processor, the steps of a method for evaluating the winding state of a synchronous generator according to any one of claims 1 to 3 are implemented.
