Generator rotor winding short-circuit prediction method based on weight update and related products

Abstract

The present invention discloses a generator rotor winding short-circuit prediction method based on weight update and related products, belonging to the technical field of generators. By obtaining the historical stator current sequence and inputting N mutually independent prediction models, N first prediction results are obtained; the reference sequence is compared with the first prediction result, the error coefficient is calculated, and the weight coefficient of each model is determined accordingly to obtain the final prediction result; the weight coefficient is adjusted until the error of the final prediction result is less than the set value; the prediction result is brought into equilibrium by compensating the weight coefficient, and the final weight coefficient of each prediction model is obtained. Finally, these final weight coefficients and prediction sequences are used to predict the stator current sequence at future moments, and compared with a pre-set threshold to predict whether the generator rotor winding is short-circuited; through multi-model fusion and weight optimization, the accuracy of stator current prediction and the reliability of fault detection are improved.

Description

Technical Field

[0001] The invention belongs to the technical field of generators, and relates to a generator rotor winding short-circuit prediction method based on weight updating and related products. Background Art

[0002] In the power system, the generator is the core equipment for power conversion, and its stable operation is crucial to the safety and reliability of the power system. However, due to the complex working environment, long-term operation wear and other reasons, the generator rotor winding may have a short circuit fault, which will not only affect the normal operation of the generator, but also may cause serious harm to the entire power system. Therefore, it is particularly important to predict and diagnose the generator rotor winding short circuit fault.

[0003] With the continuous development of power technology and the improvement of intelligence level, generator fault diagnosis technology has also developed rapidly. At present, the generator rotor winding short-circuit prediction technology is mainly based on the online measured terminal voltage, active power, and reactive power input into the established prediction model, and the output excitation current prediction value is compared with the online measured generator excitation current. If the error exceeds the set threshold, it is diagnosed as a turn-to-turn short-circuit fault in the synchronous generator rotor winding. However, although the generator rotor winding short-circuit prediction model has achieved certain results in predicting short-circuit faults, the model itself may still be affected by various factors, resulting in certain errors in the prediction results. In order to reduce this error, some research literatures have proposed the use of a combined model method, that is, by fusing the output results of multiple single prediction models, in order to obtain more accurate prediction results.

[0004] However, the core challenge of the combined model is how to reasonably allocate the weight coefficients of each single model in the combination; different prediction models have different performances, and the prediction results of a single model often have greater uncertainty. Therefore, how to dynamically adjust the weight coefficients of each single model to obtain more accurate and stable prediction results has become a technical problem that needs to be solved in the current field of generator rotor winding short-circuit prediction. Summary of the invention

[0005] The present invention provides a generator rotor winding short-circuit prediction method, system, device and medium based on weight updating, so as to solve the problem that the weight coefficients of each single model in the existing combined model are improperly distributed, resulting in reduced prediction accuracy of the combined model, thereby reducing the accuracy of generator rotor winding short-circuit prediction and the reliability of fault detection.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] In a first aspect, the present invention provides a method for predicting a generator rotor winding short circuit based on weight updating, comprising the following steps:

[0008] S1, obtaining a historical stator current sequence, wherein the historical stator current sequence includes a prediction sequence and a reference sequence;

[0009] S2, inputting the prediction sequence and the reference sequence into the improved prediction model to obtain the predicted stator current sequence at the future moment;

[0010] S3, comparing the predicted stator current sequence at the future moment with a preset threshold value to predict whether the generator rotor winding is short-circuited;

[0011] The predicted sequence and the reference sequence are input into the improved prediction model to obtain the predicted stator current sequence at the future moment, specifically:

[0012] Inputting the prediction sequence into N mutually independent prediction models in sequence to obtain corresponding N first prediction results;

[0013] Compare the reference sequence with each first prediction result to obtain an error coefficient corresponding to each prediction model;

[0014] According to the error coefficient corresponding to each prediction model, the weight coefficient corresponding to each prediction model is obtained;

[0015] Based on the first prediction result of each prediction model and the weight coefficient corresponding to each prediction model, a final prediction result is obtained;

[0016] The weight coefficient corresponding to each prediction model is adjusted until the final error corresponding to the final prediction result is less than the set value, and a new final prediction result is obtained;

[0017] Based on the new final prediction result, the weight coefficient corresponding to each prediction model is compensated until the final prediction result is in a balanced state, and the final weight coefficient corresponding to each prediction model is obtained;

[0018] Based on the final weight coefficient corresponding to each prediction model, the prediction sequence is input into N independent prediction models in sequence to obtain the predicted stator current sequence at the future moment.

[0019] Furthermore, the prediction model is a GRU model, a BP model or an ELM model.

[0020] Furthermore, the error coefficient is the sum of squares of errors between each first prediction result and the reference sequence.

[0021] Furthermore, the weight coefficient corresponding to each prediction model is:

[0022]

[0023] in, For the i The weight coefficient corresponding to the prediction model is For the i The error coefficient corresponding to the prediction model is for N The error coefficient series corresponding to the prediction model is express N The maximum value of the error coefficient series corresponding to the prediction model, Indicates arrive The sum of the error coefficients corresponding to the prediction models.

[0024] Furthermore, the specific method for obtaining the final prediction result based on the first prediction result of each prediction model and the weight coefficient corresponding to each prediction model is:

[0025] Multiplying the first prediction result of each prediction model by the weight coefficient corresponding to each prediction model to obtain a second prediction result;

[0026] Based on the second prediction result, a final prediction result is obtained; the final prediction result is:

[0027]

[0028] in, For the i The first prediction result of the prediction model, For the i The weight coefficient corresponding to the prediction model.

[0029] Furthermore, the weight coefficient corresponding to each prediction model is adjusted until the final error corresponding to the final prediction result is less than the set value, thereby obtaining a new final prediction result, which specifically includes:

[0030] Calculate the final error corresponding to the final prediction result;

[0031] When the final error is greater than the set value, the weight coefficient corresponding to each prediction model is adjusted until the final error is less than or equal to the set value:

[0032]

[0033] in, For the i The prediction model is t The weight coefficient corresponding to the adjustment; for N The prediction model is j The weight coefficient sequence corresponding to the adjustment; Indicates 1 to t A random number between ; express N The prediction model is j The minimum value of the weight coefficient sequence corresponding to the adjustment; express N The prediction model is j The maximum value of the weight coefficient sequence corresponding to the adjustment; Indicates the number of adjustments t The sine value of ; Indicates arrive Adjusted sum;

[0034] Based on the adjusted weight coefficient, the final prediction result is recalculated to obtain a new final prediction result.

[0035] Furthermore, based on the new final prediction result, the weight coefficient corresponding to each prediction model is compensated until the final prediction result is in a balanced state, and the final weight coefficient corresponding to each prediction model is obtained, which specifically includes:

[0036] Based on the new final prediction result, determine whether the new final prediction result belongs to the equilibrium state; the equilibrium coefficient F for:

[0037]

[0038] in, is the maximum value of the new final prediction result, is the second largest value of the final prediction result, is the third largest value of the new final prediction result;

[0039] when F When the median of the new final prediction result is greater than 6 times, the new final prediction result is in a non-equilibrium state, and the weight corresponding to each prediction model is compensated until the new final prediction result is in an equilibrium state. F When it is less than or equal to 6 times the median of the new final prediction result, it is in equilibrium; the specific method of compensating the weight corresponding to each prediction model is:

[0040]

[0041] in, It is i The prediction sequence of the input corresponding to the prediction model, For the i The prediction model corresponds to t The weight coefficient after adjustment is For the i The minimum value of the input prediction sequence corresponding to the prediction model, express e of Second power, e is the Euler number, ln() is e The logarithmic function with base .

[0042] In a second aspect, the present invention provides a storage medium storing a computer program implemented by the above-mentioned generator rotor winding short-circuit prediction method based on weight update, and when the computer program is called and run by a computer, the steps included in the method are executed.

[0043] In a third aspect, the present invention provides a device including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0044] In a fourth aspect, the present invention provides a generator rotor winding short-circuit prediction system based on weight update, based on the above-mentioned generator rotor winding short-circuit prediction method based on weight update, comprising:

[0045] A data acquisition module, used to acquire a historical stator current sequence, wherein the historical stator current sequence includes a prediction sequence and a reference sequence;

[0046] A prediction module, used for inputting the prediction sequence and the reference sequence into the improved prediction model to obtain the predicted stator current sequence at the future moment;

[0047] A judgment module is used to compare the predicted stator current sequence at a future moment with a preset threshold value to predict whether the generator rotor winding is short-circuited;

[0048] The predicted sequence and the reference sequence are input into the improved prediction model to obtain the predicted stator current sequence at the future moment, specifically:

[0049] Inputting the prediction sequence into N mutually independent prediction models in sequence to obtain corresponding N first prediction results;

[0050] Compare the reference sequence with each first prediction result to obtain an error coefficient corresponding to each prediction model;

[0051] According to the error coefficient corresponding to each prediction model, the weight coefficient corresponding to each prediction model is obtained;

[0052] Based on the first prediction result of each prediction model and the weight coefficient corresponding to each prediction model, a final prediction result is obtained;

[0053] The weight coefficient corresponding to each prediction model is adjusted until the final error corresponding to the final prediction result is less than the set value, and a new final prediction result is obtained;

[0054] Based on the new final prediction result, the weight coefficient corresponding to each prediction model is compensated until the final prediction result is in a balanced state, and the final weight coefficient corresponding to each prediction model is obtained;

[0055] Based on the final weight coefficient corresponding to each prediction model, the prediction sequence is input into N independent prediction models in sequence to obtain the predicted stator current sequence at the future moment.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] The present invention predicts the same historical stator current sequence by using multiple independent prediction models, and dynamically adjusts its weight coefficient based on the error coefficient of the prediction results of each prediction model. This method effectively integrates the prediction advantages of multiple models and reduces the deviation that may be caused by a single model; in the prediction process, the model weight coefficient is continuously adjusted until the final prediction error meets the set standard. This process not only ensures the accuracy of the prediction results, but also enhances the adaptability of the prediction model to the operating state of the generator; even in the face of different operating conditions or data feature changes, the model can maintain a high prediction performance through weight adjustment. Through the weight coefficient compensation mechanism, it is ensured that while meeting the prediction accuracy requirements, the weight coefficients of each prediction model reach a balance state; this balance not only helps to avoid the overfitting problem caused by excessive weights of some models, but also improves the stability and reliability of the entire prediction system; by comparing the predicted future moment stator current sequence with the preset threshold, the method can timely discover the risk of short circuit of the generator rotor winding and provide sufficient warning time for the operation and maintenance personnel; this helps to take necessary preventive measures to reduce downtime and economic losses caused by sudden failures.

[0058] Furthermore, by implementing the method of the present invention as a computer program and storing it in a specific storage medium, the method can be easily deployed and updated in various computer systems; operation and maintenance personnel can call the program through simple operations to achieve real-time monitoring and prediction of the risk of short-circuit of the generator rotor winding; and the method is easy to deploy and maintain. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0060] Figure 1 It is a block diagram of a generator rotor winding short-circuit prediction method based on weight update of the present invention;

[0061] Figure 2 It is a flow chart of the generator rotor winding short circuit prediction method based on weight update of the present invention;

[0062] Figure 3 This is a block diagram of the generator rotor winding short-circuit prediction system based on weight updating of the present invention.

[0063] Among them, 301-data acquisition module; 302-prediction module; 303-judgment module. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention described and marked in the drawings here can be arranged and designed in various different configurations.

[0065] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0066] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0067] In the description of the embodiments of the present invention, it should be noted that if the terms "upper", "lower", "horizontal", "inner", etc. indicate an orientation or positional relationship based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship in which the product of the invention is usually placed when in use, it is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0068] In addition, if the term "horizontal" appears, it does not mean that the component must be absolutely horizontal, but can be slightly tilted. For example, "horizontal" only means that its direction is more horizontal than "vertical", which does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0069] In the description of the embodiments of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "set", "install", "connect", and "connect" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal connection of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0070] Terminology explanation:

[0071] 1. GRU, the Chinese name is Gated Recurrent Unit, and the English name is Gated Recurrent Unit.

[0072] 2. BP, the Chinese name is Back Propagation, and the English name is Back Propagation.

[0073] 3. ELM, the Chinese name is Extreme Learning Machine, and the English name is Extreme Learning Machine.

[0074] 4. SSE, the Chinese name is the sum of squared errors, and the English name is Sum of Squared Errors.

[0075] 5. MAPE, the Chinese name is Mean Absolute Percentage Error, and the English name is Mean Absolute Percentage Error.

[0076] 6. HDD, the Chinese name is Hard Disk Drive, and the English name is Hard-Disk Drive.

[0077] 7. SSD, the Chinese name is Solid State Disk, and the English name is Solid State Disk.

[0078] 8. USB, the Chinese name is flash drive, and the English name is Universal Serial Bus.

[0079] 9. CD-ROM, the Chinese name is read-only compact disc, and the English name is Compact Disc Read-Only Memory.

[0080] 10. DVD, the Chinese name is Digital Video Disc, and the English name is Digital Video Disc.

[0081] The present invention is further described in detail below in conjunction with the accompanying drawings:

[0082] See also Figure 1 The present invention discloses a method for predicting short circuit of a generator rotor winding based on weight updating, comprising the following steps:

[0083] S1, obtaining a historical stator current sequence, wherein the historical stator current sequence includes a prediction sequence and a reference sequence, and the prediction sequence is used to predict the reference sequence;

[0084] First, historical stator current sequence data are obtained from the generator monitoring system; these data usually include a timestamp and a corresponding stator current value; the obtained historical stator current sequence data are sorted in chronological order and divided into a prediction sequence and a reference sequence.

[0085] S2, input the predicted sequence and the reference sequence into the improved prediction model to obtain the predicted stator current sequence at the future moment.

[0086] S201, inputting the prediction sequence into N mutually independent prediction models in sequence, predicting the reference sequence, and obtaining corresponding N first prediction results;

[0087] According to actual needs, select N independent prediction models for training. In this embodiment, you can select a GRU (Gated Recurrent Unit) model, a BP (Back Propagation) neural network model, or an ELM (Extreme Learning Machine) model. These models have shown good performance in time series prediction; use the prediction sequence to train the selected N models.

[0088] The preprocessed prediction sequence is sequentially input into the trained N prediction models to obtain N first prediction results; each prediction result is a prediction of the stator current value within a future period of time.

[0089] S202, comparing the reference sequence with each first prediction result to obtain an error coefficient corresponding to each prediction model;

[0090] The reference sequence is compared with each first prediction result, and the error coefficient of each prediction model is calculated.

[0091] S203, obtaining a weight coefficient corresponding to each prediction model according to the error coefficient corresponding to each prediction model;

[0092] According to the error coefficient of each prediction model, the weight coefficient of each prediction model is calculated. This method aims to balance the prediction accuracy and contribution of each prediction model to ensure the accuracy and stability of the final prediction results.

[0093] S204, obtaining a final prediction result based on the first prediction result of each prediction model and the weight coefficient corresponding to each prediction model;

[0094] The first prediction result of each prediction model is multiplied by the corresponding weight coefficient to obtain the second prediction result. N The second prediction results are weighted and summed to obtain the final prediction result.

[0095] S205, adjusting the weight coefficient corresponding to each prediction model until the final error corresponding to the final prediction result is less than the set value, thereby obtaining a new final prediction result;

[0096] The weight coefficients initially allocated in S203 are adjusted to further reduce the error of the final prediction result. During the adjustment process, an iterative optimization method is used to gradually adjust the weight coefficients of each prediction model until the error of the final prediction result is less than the set value. The adjustment strategy can be designed according to actual needs.

[0097] S206, based on the new final prediction result, the weight coefficient corresponding to each prediction model is compensated until the final prediction result is in a balanced state, thereby obtaining the final weight coefficient corresponding to each prediction model;

[0098] Based on the adjusted weight coefficient and the new final prediction result, weight compensation is performed to ensure that the final prediction result is in a balanced state. First, the balance coefficient is calculated to determine whether the final prediction result is balanced. If it is not balanced, the weight is compensated until the final prediction result reaches a balanced state. The compensation strategy can be designed according to the actual situation.

[0099] S207, based on the final weight coefficient corresponding to each prediction model, the prediction sequence is sequentially input into N mutually independent prediction models to obtain a predicted stator current sequence at a future moment;

[0100] The prediction sequence is predicted using the final determined weight coefficient and prediction model to obtain the stator current sequence at future moments.

[0101] S3, comparing the predicted stator current sequence at the future moment with a preset threshold value to predict whether the generator rotor winding is short-circuited;

[0102] The stator current sequence at the future moment is compared with the preset threshold; if the predicted value exceeds the threshold, it is judged that there may be a short circuit risk in the generator rotor winding; otherwise, it is considered that the generator is operating normally.

[0103] See also Figure 2 , which is a preferred embodiment of the present invention, discloses a generator rotor winding short circuit prediction method based on weight update, comprising the following steps:

[0104] S1, obtaining a historical stator current sequence, wherein the historical stator current sequence includes a prediction sequence and a reference sequence, and the prediction sequence is used to predict the reference sequence;

[0105] First, historical stator current sequence data are obtained from the generator monitoring system; these data usually include a timestamp and a corresponding stator current value; the obtained historical stator current sequence data are sorted in chronological order and divided into a prediction sequence and a reference sequence; in this embodiment, the first 90% of the historical stator current sequence is selected as the prediction sequence, and the last 10% is selected as the reference sequence.

[0106] S2, inputting the prediction sequence into N mutually independent prediction models in sequence, predicting the reference sequence, and obtaining corresponding N first prediction results;

[0107] According to actual needs, select N independent prediction models for training. In this embodiment, you can select a GRU (Gated Recurrent Unit) model, a BP (Back Propagation) neural network model, or an ELM (Extreme Learning Machine) model. These models have shown good performance in time series prediction; use the prediction sequence to train the selected N models.

[0108] The preprocessed prediction sequence is input into the trained N models in sequence to obtain N first prediction results; each prediction result is a prediction of the stator current value within a period of time in the future.

[0109] S3, comparing the reference sequence with each first prediction result to obtain an error coefficient corresponding to each prediction model;

[0110] The reference sequence is compared with each first prediction result, and the error coefficient of each prediction model is calculated. In this embodiment, the sum of squared errors (SSE) is used as the error measurement indicator. The smaller the SSE (Sum of Squared Errors), the higher the prediction accuracy of the model.

[0111] S4, obtaining a weight coefficient corresponding to each prediction model according to the error coefficient corresponding to each prediction model;

[0112] According to the error coefficient of each prediction model, the weight coefficient of each prediction model is calculated. This method aims to balance the prediction accuracy and contribution of each prediction model to ensure the accuracy and stability of the final prediction result; the calculation method is:

[0113]

[0114] in, For the i The weight coefficient corresponding to the prediction model is For the i The error coefficient corresponding to the prediction model is for N The error coefficient series corresponding to the prediction model is express N The maximum value of the error coefficient series corresponding to the prediction model, Indicates arrive The sum of the error coefficients corresponding to the prediction models.

[0115] S5, obtaining a final prediction result based on the first prediction result of each prediction model and the weight coefficient corresponding to each prediction model;

[0116] The first prediction result of each prediction model is multiplied by the corresponding weight coefficient to obtain the second prediction result. N The second prediction results are summed to obtain the final prediction result; the final prediction result is expressed as:

[0117]

[0118] in, For the i The first prediction result of the model is For the i The weight coefficient corresponding to each model.

[0119] S6, adjusting the weight coefficient corresponding to each prediction model until the final error corresponding to the final prediction result is less than the set value, thereby obtaining a new final prediction result;

[0120] The weight coefficients initially assigned by S4 are adjusted to further reduce the error of the final prediction result. During the adjustment process, the iterative optimization method is used to gradually adjust the weight coefficients of each prediction model until the error of the final prediction result is less than the set value; the set value of this embodiment is 13; the adjustment strategy is:

[0121] S601, according to S4, the initial weight coefficient is obtained, and the number of times the weight coefficient is adjusted t Initialized to t=1.

[0122] S602, calculating the final error corresponding to the final prediction result, and the final error is measured by Mean Absolute Percentage Error (MAPE).

[0123] S603, when the final error is greater than the set value, t Add 1 and adjust the weight coefficient corresponding to each prediction model until the final error is less than or equal to the set value, specifically:

[0124]

[0125] in, For the i The prediction model is t The weight coefficient corresponding to the adjustment; Perform the first j The weight coefficient sequence corresponding to the adjustment; Indicates 1 to t A random number between ; express N The prediction model is j The minimum value of the weight coefficient sequence corresponding to the adjustment; express N The prediction model is j The maximum value of the weight coefficient sequence corresponding to the adjustment; Indicates the number of adjustments t The sine value of ; Indicates arrive Adjusted sum.

[0126] S604: recalculate the final prediction result based on the adjusted weight coefficient to obtain a new final prediction result.

[0127] S7, based on the new final prediction result, the weight coefficient corresponding to each prediction model is compensated until the final prediction result is in a balanced state, thereby obtaining the final weight coefficient corresponding to each prediction model;

[0128] Based on the adjusted weight coefficient and the new final prediction result, weight compensation is performed to ensure that the final prediction result is in a balanced state. First, the balance coefficient is calculated to determine whether the final prediction result is balanced. If it is not balanced, the weight is compensated until the final prediction result reaches a balanced state. The balance coefficient is:

[0129] in, is the maximum value of the new final prediction result, is the second largest value of the final prediction result, is the third largest value of the new final prediction result.

[0130] when F When it is greater than 6 times the median of the new final prediction result, the new final prediction result is in a non-equilibrium state, and the weight corresponding to each prediction model is compensated until the new final prediction result is in an equilibrium state; F When it is less than or equal to 6 times the median of the new final prediction result, it is in equilibrium; the specific method of compensating the weight corresponding to each prediction model is:

[0131]

[0132] in, It is i The prediction sequence of the input corresponding to the prediction model, For the i The prediction model corresponds to t The weight coefficient after adjustment is For the i The minimum value of the input prediction sequence corresponding to the prediction model, express e of Second power, e is the Euler number, ln() is e The logarithmic function with base .

[0133] S8, based on the final weight coefficient corresponding to each prediction model, the prediction sequence is sequentially input into N mutually independent prediction models to obtain the predicted stator current sequence at the future moment;

[0134] The prediction sequence is predicted using the final determined weight coefficient and prediction model to obtain the stator current sequence at future moments.

[0135] S9, comparing the predicted stator current sequence at the future moment with a preset threshold value to predict whether the generator rotor winding is short-circuited;

[0136] The stator current sequence at the future moment is compared with the preset threshold; if the predicted value exceeds the threshold, it is judged that there may be a short circuit risk in the generator rotor winding; otherwise, it is considered that the generator is operating normally.

[0137] In order to further verify the advantages of the present invention, we designed a comparative experiment to compare the generator rotor winding short-circuit prediction method based on weight update proposed in the present invention (hereinafter referred to as "the present method") with the traditional prediction method (simple average model prediction). In the experiment, we selected two groups of representative stator current sequences (stator current sequence 1 and stator current sequence 2) as test objects, and used the mean absolute percentage error (MAPE, Mean Absolute Percentage Error) as the key indicator to measure the prediction accuracy. The results are as follows:

[0138]

[0139] It can be seen from the experimental results that the method of the present invention has significantly improved the prediction accuracy compared with the traditional model. In the two sets of stator current sequences, MAPE (Mean Absolute Percentage Error) decreased by 50.4% and 54.7% respectively, indicating that the method of the present invention can more accurately capture the changing trend of the generator stator current and improve the accuracy of the prediction. At the same time, the weight adjustment and compensation steps play an important role in further improving the prediction accuracy; by dynamically adjusting the weights of each model, the method of the present invention can better adapt to the changing characteristics of the data and ensure that the prediction results always maintain a high level of accuracy.

[0140] The present invention also discloses a storage medium, which stores a computer program implemented according to the above method. When the computer program is called and run by a computer, the steps included in the method are executed; the storage medium of the present invention is a non-volatile or readable and writable data storage device, which is designed to store and retain computer program instructions. The storage medium can be, but is not limited to, a hard disk drive (HDD, Hard-Disk Drive), a solid state drive (SSD, Solid State Disk), a flash drive (USB, Universal Serial Bus), an optical disk (such as CD-ROM (Compact Disc Read-Only Memory, read-only optical disk), DVD (Digital Video Disc, digital video optical disk)), a virtual storage space in a cloud storage service, or any other medium that can store and allow a computer to access and execute program code.

[0141] The present invention discloses a device, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program; the device can be any device with computing power, such as a personal computer, a workstation, a server, an embedded system, an intelligent terminal, etc. The device can be deployed in the monitoring center of a power plant to monitor the operating status of the generator in real time, predict and warn of potential rotor winding short-circuit faults; through network connection, remote monitoring and predictive analysis can be realized to improve operation and maintenance efficiency and response speed; integrated into the data center management system as part of the power system fault prediction, the overall operation and maintenance level can be improved.

[0142] See also Figure 3 The present invention discloses a generator rotor winding short circuit prediction system based on weight update, comprising:

[0143] A data acquisition module 301 is used to acquire a historical stator current sequence, wherein the historical stator current sequence includes a prediction sequence and a reference sequence;

[0144] A prediction module 302 is used to input the prediction sequence and the reference sequence into the improved prediction model to obtain the predicted stator current sequence at a future moment;

[0145] The judgment module 303 is used to compare the predicted stator current sequence at a future moment with a preset threshold value to predict whether the generator rotor winding is short-circuited;

[0146] The predicted sequence and the reference sequence are input into the improved prediction model to obtain the predicted stator current sequence at the future moment, specifically:

[0147] Inputting the prediction sequence into N mutually independent prediction models in sequence to obtain corresponding N first prediction results;

[0148] Compare the reference sequence with each first prediction result to obtain an error coefficient corresponding to each prediction model;

[0149] According to the error coefficient corresponding to each prediction model, the weight coefficient corresponding to each prediction model is obtained;

[0150] Based on the first prediction result of each prediction model and the weight coefficient corresponding to each prediction model, a final prediction result is obtained;

[0151] The weight coefficient corresponding to each prediction model is adjusted until the final error corresponding to the final prediction result is less than the set value, and a new final prediction result is obtained;

[0152] Based on the new final prediction result, the weight coefficient corresponding to each prediction model is compensated until the final prediction result is in a balanced state, and the final weight coefficient corresponding to each prediction model is obtained;

[0153] Based on the final weight coefficient corresponding to each prediction model, the prediction sequence is input into N independent prediction models in sequence to obtain the predicted stator current sequence at the future moment.

[0154] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for predicting short circuit of generator rotor winding based on weight updating, characterized in that: The following steps are involved: Acquire a historical stator current sequence, wherein the historical stator current sequence includes a prediction sequence and a reference sequence; The prediction sequence and the reference sequence are input into the improved prediction model to obtain the predicted stator current sequence at the future moment; Compare the predicted stator current sequence at the future moment with the preset threshold value to predict whether the generator rotor winding is short-circuited; The predicted sequence and the reference sequence are input into the improved prediction model to obtain the predicted stator current sequence at the future moment, specifically: Inputting the prediction sequence into N mutually independent prediction models in sequence to obtain corresponding N first prediction results; Compare the reference sequence with each first prediction result to obtain an error coefficient corresponding to each prediction model; According to the error coefficient corresponding to each prediction model, the weight coefficient corresponding to each prediction model is obtained; Based on the first prediction result of each prediction model and the weight coefficient corresponding to each prediction model, a final prediction result is obtained; The weight coefficient corresponding to each prediction model is adjusted until the final error corresponding to the final prediction result is less than the set value, and a new final prediction result is obtained; Based on the new final prediction result, the weight coefficient corresponding to each prediction model is compensated until the final prediction result is in a balanced state, and the final weight coefficient corresponding to each prediction model is obtained; Based on the final weight coefficient corresponding to each prediction model, the prediction sequence is input into N independent prediction models in sequence to obtain the predicted stator current sequence at the future moment.

2. The generator rotor winding short circuit prediction method based on weight update according to claim 1 is characterized in that: The prediction model is a GRU model, a BP model or an ELM model.

3. The generator rotor winding short circuit prediction method based on weight update according to claim 1 is characterized in that: The error coefficient is the sum of squares of errors between each first prediction result and the reference sequence.

4. The generator rotor winding short circuit prediction method based on weight update according to claim 1 is characterized in that: The weight coefficient corresponding to each prediction model is: in, For the i The weight coefficient corresponding to the prediction model is For the i The error coefficient corresponding to the prediction model is for N The error coefficient series corresponding to the prediction model is express N The maximum value of the error coefficient series corresponding to the prediction model, Indicates arrive The sum of the error coefficients corresponding to the prediction models.

5. The generator rotor winding short circuit prediction method based on weight update according to claim 1 is characterized in that: The first prediction result based on each prediction model and the weight coefficient corresponding to each prediction model are used to obtain the final prediction result. The specific method is as follows: Multiplying the first prediction result of each prediction model by the weight coefficient corresponding to each prediction model to obtain a second prediction result; Based on the second prediction result, a final prediction result is obtained; the final prediction result is: in, For the i The first prediction result of the prediction model, For the i The weight coefficient corresponding to the prediction model.

6. The generator rotor winding short circuit prediction method based on weight update according to claim 1 is characterized in that: The weight coefficient corresponding to each prediction model is adjusted until the final error corresponding to the final prediction result is less than the set value, thereby obtaining a new final prediction result, which specifically includes: Calculate the final error corresponding to the final prediction result; When the final error is greater than the set value, the weight coefficient corresponding to each prediction model is adjusted until the final error is less than or equal to the set value: in, For the i The prediction model is t The weight coefficient corresponding to the adjustment; for N The prediction model is j The weight coefficient sequence corresponding to the adjustment; Indicates 1 to t A random number between ; express N The prediction model is j The minimum value of the weight coefficient sequence corresponding to the adjustment; express N The prediction model is j The maximum value of the weight coefficient sequence corresponding to the adjustment; Indicates the number of adjustments t The sine value of ; Indicates arrive Adjusted sum; Based on the adjusted weight coefficient, the final prediction result is recalculated to obtain a new final prediction result.

7. The generator rotor winding short circuit prediction method based on weight update according to claim 1 is characterized in that: Based on the new final prediction result, the weight coefficient corresponding to each prediction model is compensated until the final prediction result is in a balanced state, and the final weight coefficient corresponding to each prediction model is obtained, which specifically includes: Based on the new final prediction result, determine whether the new final prediction result belongs to the equilibrium state; the equilibrium coefficient F for: in, is the maximum value of the new final prediction result, is the second largest value of the final prediction result, is the third largest value of the new final prediction result; when F When the median of the new final prediction result is greater than 6 times, the new final prediction result is in a non-equilibrium state, and the weight corresponding to each prediction model is compensated until the new final prediction result is in an equilibrium state. F When it is less than or equal to 6 times the median of the new final prediction result, it is in equilibrium; the specific method of compensating the weight corresponding to each prediction model is: in, It is i The prediction sequence of the input corresponding to the prediction model, For the i The prediction model corresponds to t The weight coefficient after adjustment is For the i The minimum value of the input prediction sequence corresponding to the prediction model, express e of Second power, e is the Euler number, ln() is e The logarithmic function with base .

8. A storage medium, characterized in that: It stores a computer program implemented according to the generator rotor winding short-circuit prediction method based on weight update as described in any one of claims 1 to 7, and the computer program executes the method when it is called and run by a computer.

9. A device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the generator rotor winding short-circuit prediction method based on weight updating according to any one of claims 1 to 7 is implemented.

10. A generator rotor winding short-circuit prediction system based on weight update, based on the generator rotor winding short-circuit prediction method based on weight update according to claim 1, characterized in that: include: A data acquisition module, used to acquire a historical stator current sequence, wherein the historical stator current sequence includes a prediction sequence and a reference sequence; A prediction module, used for inputting the prediction sequence and the reference sequence into the improved prediction model to obtain the predicted stator current sequence at the future moment; A judgment module is used to compare the predicted stator current sequence at a future moment with a preset threshold value to predict whether the generator rotor winding is short-circuited; The predicted sequence and the reference sequence are input into the improved prediction model to obtain the predicted stator current sequence at the future moment, specifically: Inputting the prediction sequence into N mutually independent prediction models in sequence to obtain corresponding N first prediction results; Compare the reference sequence with each first prediction result to obtain an error coefficient corresponding to each prediction model; According to the error coefficient corresponding to each prediction model, the weight coefficient corresponding to each prediction model is obtained; Based on the first prediction result of each prediction model and the weight coefficient corresponding to each prediction model, a final prediction result is obtained; The weight coefficient corresponding to each prediction model is adjusted until the final error corresponding to the final prediction result is less than the set value, and a new final prediction result is obtained; Based on the new final prediction result, the weight coefficient corresponding to each prediction model is compensated until the final prediction result is in a balanced state, and the final weight coefficient corresponding to each prediction model is obtained; Based on the final weight coefficient corresponding to each prediction model, the prediction sequence is input into N independent prediction models in sequence to obtain the predicted stator current sequence at the future moment.

Images