Water-turbine generator set stator bar residual life evaluation method based on self-adaptive working conditions

Abstract

The invention discloses a method for evaluating the residual life of a stator bar of a water-turbine generator set based on self-adaptive working conditions. The method comprises the following steps: S1, data acquisition: acquiring operation working condition data of a water-turbine generator in real time; s2, working condition identification: working condition interval division based on clustering and a fuzzy membership function is adopted; s3, extracting life characteristics; s4, model establishment and training: constructing a residual life prediction model by adopting a hybrid intelligent algorithm combining an improved particle swarm IPSO and a BP neural network, and then training an optimal prediction model under each working condition; s5, performing prediction feedback; according to the method, the insulation state of the stator bar can be accurately evaluated, so that the effective residual life of the stator bar can be predicted, a power enterprise can carry out targeted maintenance on the bar according to a prediction result, the non-planned downtime of a unit is reduced, a scientific basis is provided for operation decision making of the unit, and the maximum economic benefit is fully realized.

Description

Technical Field

[0001] This invention relates to the field of residual life assessment technology for stator bars of hydro-generators, and in particular to a method for assessing the residual life of stator bars of hydro-generator sets based on adaptive operating conditions. Background Technology

[0002] Generators are the core equipment of hydropower stations, and generator stator insulation is the weakest link and a critical factor in generator performance. Furthermore, insulation failures often result in severe losses. During operation, stator bars are subjected to electrical, thermal, and mechanical forces, causing their insulation strength to gradually decrease or even be lost, leading to insulation breakdown. This results in grounding, short circuits, and other phenomena in the windings, seriously affecting the safe and stable operation of the generator and shortening its service life. According to statistics from relevant organizations, 50% of generator failures originate from insulation problems, and 70% of these insulation-related failures cause unplanned downtime exceeding 50 days. Moreover, insulation failures cause the most severe losses among all types of generator failures.

[0003] The actual operating conditions of a hydro turbine unit can change due to various factors, such as operating head, active load, and guide vane opening. These changes in operating conditions cause the stator bars to be subjected to different stresses, temperatures, and other conditions, which in turn leads to changes in their insulation capacity and their effective residual life. Traditional life assessment methods are difficult to adapt to the dynamic changes in operating conditions, resulting in reduced accuracy and reliability of assessment results.

[0004] Given the widespread corona and electro-corrosion phenomena in the stator bars of hydropower stations, the decline in the insulation capacity of the stator bars is a cause for concern. Therefore, it is necessary to utilize monitoring data to explore their insulation characteristics, make accurate assessments of their insulation status, and predict their effective remaining life. This will enable power companies to carry out targeted maintenance on the stator bars based on the prediction results, reduce unplanned downtime of the units, provide a scientific basis for unit operation decisions, and fully realize maximum economic benefits. Summary of the Invention

[0005] The purpose of this invention is to overcome the above-mentioned shortcomings and provide a method for assessing the residual life of stator bars in hydro-generator units based on adaptive operating conditions. This method can accurately assess the insulation status of stator bars, thereby predicting their effective remaining life. This allows power companies to perform targeted maintenance on the bars based on the prediction results, reducing unplanned downtime of the unit, providing a scientific basis for unit operation decisions, and maximizing economic benefits.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a method for assessing the residual life of stator bars of a hydro-generator unit based on adaptive operating conditions, which includes the following steps:

[0007] S1. Data Acquisition: Real-time acquisition of operating condition data of the hydro-generator;

[0008] S2. Operating condition identification: Operating condition intervals are divided based on clustering and fuzzy membership functions;

[0009] S3. Lifetime feature extraction;

[0010] S4. Model building and training: A hybrid intelligent algorithm combining improved particle swarm optimization (IPSO) and backpropagation (BP) neural network is used to build the remaining lifetime prediction model, and then the optimal prediction model under each working condition is trained.

[0011] S5, Predictive Feedback.

[0012] Preferably, step S1 specifically includes the following process: real-time acquisition of operating condition data of the hydro-generator, including head, flow rate, load, temperature of stator bars, and stress parameters.

[0013] Preferably, the operating condition data is collected through a field monitoring sensor network and transmitted to the data processing center.

[0014] Preferably, step S2 specifically includes the following process:

[0015] S2.1 Preliminary partitioning based on clustering:

[0016] First, the historical operation dataset is normalized to eliminate the influence of dimensions, with head, opening degree, and load as input features and bar temperature as clustering constraint. Then, a Gaussian Mixture Model (GMM) is trained on the standardized data to obtain the mean and covariance matrix of each Gaussian distribution. For any sample x = (h, p, k), where h represents head, p represents power, and k represents opening degree, the probability of it belonging to the i-th operating condition is calculated.

[0017] Then, for each cluster, the boundaries of its head, power, and aperture are calculated to form a preliminary set of operating conditions:

[0018] Cset = {C1, C2, ..., C} n}

[0019] Among them, each working condition C i Corresponding to a parameter range:

[0020] C i =[{hmin,hmax},{pmin,pmax},{kmin,kmax}]

[0021] In the above formula, hmin and hmax represent operating condition C. i The corresponding head range, {pmin, pmax}, represents operating condition C. iThe corresponding power range, {kmin, kmax} represents operating condition C. i The corresponding range of opening degree;

[0022] S2.2 Optimization based on fuzzy membership function:

[0023] First, for each working condition C i Define the fuzzy membership functions for its head, power, and opening degree;

[0024] Then, for any running point x = (h, p, k) in the dataset, calculate its belonging to each working condition C. i Overall membership degree:

[0025] μi(x)=wH·μH(h)+wP·μP(p)+wK·μK(k)

[0026] Wherein, the weights of parameters wH, wP, and wK are given, and μH(h), μP(p), and μK(k) represent the univariate membership functions of each dimension, used to measure the data point's association with working condition C in a certain dimension. i The degree of membership;

[0027] The following formula represents the set of operating conditions categorized based on temperature distribution obtained from historical operating data:

[0028]

[0029] Among them, C1, C2, ... C n The specific operating conditions obtained from clustering are represented by [{hmin,hmax},{pmin,pmax},{kmin,kmax}], which represent the operating condition ranges corresponding to different temperature clusters. [{hmin,hmax},{pmin,pmax},{kmin,kmax}] represent the corresponding operating conditions at specific temperatures, namely the head range, power range, and opening range, respectively.

[0030] Preferably, in S2.1, the probability of belonging to the i-th working condition is calculated as follows:

[0031]

[0032] In the above formula, μ i (x) is the calculated posterior probability that sample x belongs to the i-th working condition; where π i It is a mixed weight; N(x|μ i ,Σ i ) is the Gaussian distribution density function, used to calculate the likelihood value of sample x under this component, μ i Let Σ be the mean vector of the i-th component. i Let be the covariance matrix of the i-th component; This represents the weighted likelihood summation over all Gaussian components, i.e., the marginal probability of sample x in the entire GMM.

[0033] Preferably, in S2.2, the fuzzy membership function of the water head is:

[0034]

[0035] Among them, h center It is the cluster center of the head under this working condition, σ H The fuzzy range of the reaction head change.

[0036] Preferably, step S3 specifically includes the following process:

[0037] The empirical formula for the remaining life of stator bar insulation is: L = L0e -kQ Where L0 is the design life of the stator bar, k is the degradation coefficient of the bar material, and Q is the cumulative discharge amount;

[0038] Different discharge quantities will cause losses to the life of the stator bars, so the specific loss of insulation life per unit time or per single discharge event can be derived; the converted formula is: ΔL=L0(1-e -kQ ); where ΔL represents the insulation life loss caused by the discharge quantity Q. Through this conversion, the direct impact of partial discharge on insulation life can be intuitively quantified for each instance or period of time, so as to assess the life loss at different time scales.

[0039] Preferably, step S4 specifically includes the following process:

[0040] S4.1 The standard particle swarm optimization algorithm uses a linearly decreasing inertia weight strategy when adjusting the inertia weight W, that is:

[0041]

[0042] In the formula, W max and W min These are the maximum and minimum values ​​of W, respectively, where t is the current iteration step. max This represents the total number of iterations; the standard particle swarm optimization algorithm is improved by modifying W using the following formula:

[0043]

[0044] In the above formula, when t is small, W is close to W0. max To ensure the algorithm's global search capability, W decreases non-linearly as t gradually increases, thus ensuring the algorithm's local search capability.

[0045] S4.2 The IPSO-BP prediction model first uses IPSO to optimize the initial weights and thresholds of the neural network to find an optimal set of weights that minimizes the input-output error under these weights. Then, the BP algorithm is used to further fine-tune the weights to find the true global optimum.

[0046] S4.3. Based on the designed model structure, establish a life prediction model for each working condition by combining the fused historical operating data and discharge data. The input is environmental parameters such as head, load, and opening degree, and the output is life loss. The optimal prediction model for each working condition is trained by dividing the dataset and adjusting the model parameters.

[0047] Preferably, step S4.2 specifically includes the following process:

[0048] (1) Initialization: n i n is the number of input neurons. h n is the number of hidden neurons. o If the number of output neurons is given, then the dimension D of the particle swarm is:

[0049] D = n h +n o +n i ×n h +n h ×n o

[0050] (2) Set the fitness function of the particle swarm optimization, using the mean square error between the desired output and the network output as the fitness function:

[0051]

[0052] In the formula, y i and t i Let be the expected output and the network output of the i-th sample, respectively, and n be the number of network samples;

[0053] (3) Optimize the weights and thresholds of the BP network using IPSO, and use the optimized weights and thresholds as the initial weights and thresholds of the BP model, and train the network until the mean square error of the network's performance index is less than the maximum permissible error e. max .

[0054] Preferably, step S5 specifically includes the following process:

[0055] In practical applications, based on the load curves issued by the dispatch center and the hydrological system data of the power station, the data is first divided into different operating period segments according to the set of operating conditions. The divided operating period segments are then input into the life loss benchmark model. Different benchmark models are adapted according to the conditions to predict the loss life under each operating condition. Finally, the overall remaining life of the bars is obtained by combining the historical life data.

[0056] Beneficial effects of this invention:

[0057] 1. Improve assessment accuracy: By considering the impact of changes in turbine operating conditions on stator bar discharge performance, the assessment parameters and models can be adaptively and dynamically adjusted, which can more accurately reflect the residual life of stator bars under different operating conditions and improve the accuracy of assessment results.

[0058] 2. Enhanced adaptability: This method can automatically adapt to dynamic changes in operating conditions without manual intervention, and has strong adaptability and flexibility.

[0059] 3. Optimize maintenance plans: Accurate residual life assessment results help to rationally arrange the maintenance plan of hydro-generators, avoid over-maintenance or under-maintenance, reduce maintenance costs, and improve the operating efficiency and reliability of equipment.

[0060] 4. Ensure equipment safety: Timely and accurate assessment of the remaining life of the stator bars can help identify potential faults in advance and take corresponding measures to ensure the safe operation of the hydro-generator.

[0061] 5. This invention utilizes stator partial discharge monitoring data combined with unit operating data to establish residual life models under different operating conditions. Then, based on load curves and water condition information over a future period, it predicts the residual life of the stator bars. It can accurately assess the insulation status of the stator bars, thereby predicting their effective remaining life. This allows power companies to perform targeted maintenance on the bars based on the prediction results, reducing unplanned unit downtime, providing a scientific basis for unit operation decisions, and maximizing economic benefits. Attached Figure Description

[0062] Figure 1 This is a schematic diagram of the overall process for an adaptive operating condition-based method for assessing the residual life of stator bars in a hydro-generator unit.

[0063] Figure 2 This is a flowchart corresponding to step 5;

[0064] Figure 3 This is a schematic diagram illustrating the process of a method for assessing the residual life of stator bars in a hydro-generator unit based on adaptive operating conditions. Detailed Implementation

[0065] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0066] Example 1: As Figure 1 and 3 As shown, a method for assessing the residual life of stator bars in a hydro-generator unit based on adaptive operating conditions includes the following steps:

[0067] Step 1. Data Acquisition:

[0068] Real-time data collection of the hydro-generator's operating conditions, including parameters such as head, flow rate, load, stator bar temperature, and stress, is performed. This data can be collected through a network of field monitoring sensors and transmitted to a data processing center.

[0069] Step 2. Operating Condition Identification:

[0070] The degradation of stator bar insulation performance is usually manifested as enhanced discharge activity, and the discharge quantity and frequency are closely related to the operating temperature of the bars. Since temperature directly affects the dielectric properties and local electric field distribution of the insulation material, bar temperature is used as the core benchmark for operating condition classification. However, in actual operation, the relationship between the unit's operating parameters (such as head, power, and operating degree) and temperature is not a simple univariate mapping, but rather influenced by the coupling of multiple factors. Traditional fixed threshold classification methods are difficult to accurately describe the transition process of different operating conditions. Therefore, this method adopts an operating condition interval classification method based on clustering (GMM clustering) and fuzzy membership functions.

[0071] (1) Preliminary partitioning based on clustering

[0072] Considering that the relationship between the turbine unit's operating parameters and the bar temperature is continuous and nonlinear, and that there are overlapping operating conditions during the transition process, a Gaussian mixture model is used for preliminary classification.

[0073] First, the historical operating dataset (head, flow rate, load, and temperature, etc.) is normalized to eliminate the influence of dimensions. Head, flow rate, and load are used as input features, and bar temperature is used as a clustering constraint. Then, a Gaussian Mixture Model (GMM) is trained on the standardized data to obtain the mean and covariance matrices of each Gaussian distribution. For any sample x = (h, p, k), where h represents head, p represents power, and k represents flow rate, the probability of it belonging to the i-th operating condition is calculated:

[0074]

[0075] In the above formula, μ i (x) is the calculated posterior probability that sample x belongs to the i-th working condition; where π i It is a mixed weight; N(x|μ i ,Σ i ) is the Gaussian distribution density function, used to calculate the likelihood value of sample x under this component, μ i Let Σ be the mean vector of the i-th component. i Let be the covariance matrix of the i-th component; This represents the weighted likelihood summation over all Gaussian components, i.e., the marginal probability of sample x in the entire GMM;

[0076] Then, for each cluster, the boundaries of its head, power, and aperture are calculated to form a preliminary set of operating conditions:

[0077] Cset = {C1, C2, ..., C} n}

[0078] Among them, each working condition C i Corresponding to a parameter range:

[0079] C i =[{hmin,hmax},{pmin,pmax},{kmin,kmax}]

[0080] In the above formula, hmin and hmax represent operating condition C. i The corresponding head range, {pmin, pmax}, represents operating condition C. i The corresponding power range, {kmin, kmax} represents operating condition C. i The corresponding range of opening degree;

[0081] (2) Optimization based on fuzzy membership function

[0082] Since the operating parameters of hydro-turbine units typically change continuously, and there are overlapping or transitional regions between different operating conditions, a hard partitioning approach might lead to misclassification of samples near the boundaries. Therefore, a fuzzy membership function is introduced to make the partitioning of operating conditions more closely reflect the actual operating characteristics of the hydro-turbine unit.

[0083] First, for each working condition C i Define the fuzzy membership functions for its head, power, and aperture. For example, the fuzzy membership function for head is:

[0084]

[0085] Among them, h center It is the cluster center of the head under this working condition, σ H The fuzzy range of the reaction head change. The fuzzy membership functions of power and opening degree are of the same form as the above formula, except that the head h is replaced by the power p and opening degree k.

[0086] Then, for any running point x = (h, p, k) in the dataset, calculate its belonging to each working condition C. i Overall membership degree:

[0087] μi(x)=wH·μH(h)+wP·μP(p)+wK·μK(k)

[0088] Where wH, wP, and wK are the weights of each parameter, and μH(h), μP(p), and μK(k) represent the univariate membership functions of each dimension, used to measure the data point's association with working condition C in a certain dimension. i The degree of membership;

[0089] The following formula represents the set of operating conditions categorized based on temperature distribution obtained from historical operating data:

[0090]

[0091] Among them, C1, C2, ... C m The specific operating conditions obtained from clustering are represented by [{hmin,hmax},{pmin,pmax},{kmin,kmax}], which represent the operating condition ranges corresponding to different temperature clusters. [{hmin,hmax},{pmin,pmax},{kmin,kmax}] represent the corresponding operating conditions at specific temperatures, namely the head range, power range, and opening range, respectively.

[0092] Step 3. Lifetime Feature Extraction:

[0093] According to the empirical formula for the remaining life of stator bar insulation, L = L0e -kQ (L0 is the design life of the stator bar, and k is the degradation coefficient of the bar material) It can be seen that this formula can be transformed into ΔL=L0(1-e -kQ This allows us to determine the lifespan loss of the bar caused by each discharge based on partial discharge monitoring data.

[0094] According to the empirical formula for the remaining life of stator bar insulation (the following formula), different discharge quantities will cause losses to the life of stator bars. Therefore, the specific loss of insulation life per unit time or per single discharge event can be derived.

[0095] L=L0e -kQ

[0096] Where L0 is the design life of the wire rod, usually provided by the manufacturer; for epoxy mica insulation materials, the design life is generally between 25 and 30 years; k is the degradation coefficient of the wire rod material, typically between 0.001 and 0.005 pC. -1 In specific applications, the value can be determined by fitting the historical data of the unit or by accelerating aging experiments; Q is the cumulative discharge amount.

[0097] The converted formula is as follows, where ΔL represents the insulation life loss caused by the discharge quantity Q. Through this conversion, the direct impact of partial discharge on insulation life can be quantified more intuitively, making it easier to assess life loss at different time scales more flexibly.

[0098] ΔL=L0(1-e -kQ )

[0099] Step 4. Model Training:

[0100] To improve the accuracy of lifetime loss prediction, a hybrid intelligent algorithm combining improved particle swarm optimization (IPSO) and backpropagation (BP) neural networks is used for remaining lifetime prediction. Standard PSO algorithms typically employ a linearly decreasing inertia weight strategy when adjusting the inertia weight W, i.e.:

[0101]

[0102] In the formula, W max and W min These are the maximum and minimum values ​​of W, respectively, and t is the current iteration step. max This represents the total number of iterations. However, using a linear decreasing method not only affects convergence efficiency in the early stages but also in the later stages. Furthermore, as W decreases, the global search capability declines, making it prone to getting trapped in local optima. Therefore, W is modified using the following formula to improve the shortcomings of the standard particle swarm optimization algorithm:

[0103]

[0104] In the above formula, when t is small, W is close to W0. max To ensure the algorithm's global search capability, W decreases non-linearly as t gradually increases, ensuring the algorithm's local search capability and effectively avoiding the drawback of getting trapped in local optima. It can flexibly adjust the balance between global search and local search capabilities.

[0105] The IPSO-BP prediction model first uses IPSO to optimize the initial weights and thresholds of the neural network to find an optimal set of weights that minimizes the input-output error. Then, the BP algorithm is used to further fine-tune the weights to find the true global optimum. The specific process is as follows:

[0106] (1) Initialization: n i n is the number of input neurons. h n is the number of hidden neurons. o If the number of output neurons is given, then the dimension D of the particle swarm is:

[0107] D = n h +n o +n i ×n h +n h ×n o

[0108] (2) Set the fitness function of the particle swarm optimization, using the mean square error between the desired output and the network output as the fitness function:

[0109]

[0110] In the formula, y i and t i Let be the expected output and the network output of the i-th sample, respectively, and n be the number of network samples.

[0111] (3) Optimize the weights and thresholds of the BP network using IPSO, and use the optimized weights and thresholds as the initial weights and thresholds of the BP model, and train the network until the mean square error of the network's performance index is less than e. max (Maximum permissible error).

[0112] Based on the designed model structure, a life prediction model for each operating condition is established by combining the fused historical operating condition data and discharge data. The inputs are environmental parameters such as head, load, and opening degree, and the output is life loss. The optimal prediction model for each operating condition is trained by dividing the dataset and adjusting the model parameters.

[0113] Step 5. Predictive Feedback:

[0114] In practical applications, based on the load curves issued by the dispatch center and the hydrological system data of the power station, the data is first divided into different operating condition periods according to the defined set of operating conditions. These divided operating condition periods are then input into the life loss benchmark model. Different benchmark models are adaptively applied based on the conditions to predict the loss life under each operating condition. Finally, the overall remaining life of the power bars is obtained by combining historical life data. The flowchart of this process is shown below. Figure 2 As shown.

[0115] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. A method for assessing the residual life of stator bars in a hydro-generator unit based on adaptive operating conditions, characterized in that: It includes the following steps: S1. Data Acquisition: Real-time acquisition of operating condition data of the hydro-generator; S2. Operating condition identification: Operating condition intervals are divided based on clustering and fuzzy membership functions; S3. Lifetime feature extraction; S4. Model building and training: A hybrid intelligent algorithm combining improved particle swarm optimization (IPSO) and backpropagation (BP) neural network is used to build the remaining lifetime prediction model, and then the optimal prediction model under each working condition is trained. S5, Predictive Feedback.

2. The method for assessing the residual life of stator bars in a hydro-generator unit based on adaptive operating conditions according to claim 1, characterized in that: The specific steps S1 are as follows The process includes the following: real-time acquisition of operating condition data of the hydro-generator, including head, flow rate, load, stator bar temperature, and stress parameters.

3. The method for assessing the residual life of stator bars in a hydro-generator unit based on adaptive operating conditions according to claim 2, characterized in that: Operating condition data is collected through a network of on-site monitoring sensors and transmitted to the data processing center.

4. The method for assessing the residual life of stator bars in a hydro-generator unit based on adaptive operating conditions according to claim 1, characterized in that: Step S2 specifically includes the following process: S2.1 Preliminary partitioning based on clustering: First, the historical operation dataset is normalized to eliminate the influence of dimensions, with head, opening degree, and load as input features and bar temperature as clustering constraint. Then, a Gaussian Mixture Model (GMM) is trained on the standardized data to obtain the mean and covariance matrix of each Gaussian distribution. For any sample x = (h, p, k), where h represents head, p represents power, and k represents opening degree, the probability of it belonging to the i-th operating condition is calculated. Then, for each cluster, the boundaries of its head, power, and aperture are calculated to form a preliminary set of operating conditions: Cset={C1,C2,…C n } Among them, each working condition C i Corresponding to a parameter range: C i =[{hmin,hmax},{pmin,pmax},{kmin,kmax}] In the above formula, hmin and hmax represent operating condition C. i The corresponding head range, {pmin, pmax}, represents operating condition C. i The corresponding power range, {kmin, kmax} represents operating condition C. i The corresponding range of opening degree; S2.2 Optimization based on fuzzy membership function: First, for each working condition C i Define the fuzzy membership functions for its head, power, and opening degree; Then, for any running point x = (h, p, k) in the dataset, calculate its belonging to each working condition C. i Overall membership degree: μi(x)=wH·μH(h)+wP·μP(p)+wK·μK(k) Where wH, wP, and wK are the weights of each parameter, and μH(h), μP(p), and μK(k) represent the univariate membership functions of each dimension, used to measure the data point's association with working condition C in a certain dimension. i The degree of membership; The following formula represents the set of operating conditions categorized based on temperature distribution obtained from historical operating data: Among them, C1, C2, ... C n The specific operating conditions obtained from clustering are represented by [{hmin,hmax},{pmin,pmax},{kmin,kmax}], which represent the operating condition ranges corresponding to different temperature clusters. [{hmin,hmax},{pmin,pmax},{kmin,kmax}] represent the corresponding operating conditions at specific temperatures, namely the head range, power range, and opening range, respectively.

5. The method for assessing the residual life of stator bars in a hydro-generator unit based on adaptive operating conditions according to claim 4, characterized in that: In S2.1, the probability of belonging to the i-th working condition is calculated as follows: In the above formula, μ i (x) is the calculated posterior probability that sample x belongs to the i-th working condition; where π i It is a mixed weight; N(x|μ i ,Σ i ) is the Gaussian distribution density function, used to calculate the likelihood value of sample x under this component, μ i Let Σ be the mean vector of the i-th component. i Let be the covariance matrix of the i-th component; This represents the weighted likelihood summation over all Gaussian components, i.e., the marginal probability of sample x in the entire GMM.

6. The method for assessing the residual life of stator bars in a hydro-generator unit based on adaptive operating conditions according to claim 5, characterized in that: In S2.2, the fuzzy membership function of the water head is: Among them, h center It is the cluster center of the head under this working condition, σ H The fuzzy range of the reaction head change.

7. The method for assessing the residual life of stator bars in a hydro-generator unit based on adaptive operating conditions according to claim 1, characterized in that: The specific steps S3 are as follows Includes the following processes: The empirical formula for the remaining life of stator bar insulation is: L = L0e -kQ Where L0 is the design life of the stator bar, k is the degradation coefficient of the bar material, and Q is the cumulative discharge amount; Different discharge quantities will cause losses to the life of the stator bars, so the specific loss of insulation life per unit time or per single discharge event can be derived; the converted formula is: ΔL=L0(1-e -kQ ); where ΔL represents the insulation life loss caused by the discharge quantity Q. Through this conversion, the direct impact of partial discharge on insulation life can be intuitively quantified for each instance or period of time, so as to assess the life loss at different time scales.

8. The method for assessing the residual life of stator bars in a hydro-generator unit based on adaptive operating conditions according to claim 1, characterized in that: Step S4 specifically includes the following process: S4.1 The standard particle swarm optimization algorithm uses a linearly decreasing inertia weight strategy when adjusting the inertia weight W, that is: In the formula, W max and W min These are the maximum and minimum values ​​of W, respectively, where t is the current iteration step. mac This represents the total number of iterations; the standard particle swarm optimization algorithm is improved by modifying W using the following formula: In the above formula, when t is small, W is close to W0. max To ensure the algorithm's global search capability, W decreases non-linearly as t gradually increases, thus ensuring the algorithm's local search capability. S4.2 The IPSO-BP prediction model first uses IPSO to optimize the initial weights and thresholds of the neural network to find an optimal set of weights that minimizes the input-output error under these weights. Then, the BP algorithm is used to further fine-tune the weights to find the true global optimum. S4.

3. Based on the designed model structure, establish a life prediction model for each working condition by combining the fused historical operating data and discharge data. The input is environmental parameters such as head, load, and opening degree, and the output is life loss. The optimal prediction model for each working condition is trained by dividing the dataset and adjusting the model parameters.

9. The method for assessing the residual life of stator bars in a hydro-generator unit based on adaptive operating conditions according to claim 8, characterized in that: Step S4.2 specifically includes the following process: (1) Initialization: n i n is the number of input neurons. h n is the number of hidden neurons. o If the number of output neurons is given, then the dimension D of the particle swarm is: D=n h +n o +n i ×n h +n h ×n o (2) Set the fitness function of the particle swarm optimization, using the mean square error between the desired output and the network output as the fitness function: In the formula, y i and t i Let be the expected output and the network output of the i-th sample, respectively, and n be the number of network samples; (3) Optimize the weights and thresholds of the BP network using IPSO, and use the optimized weights and thresholds as the initial weights and thresholds of the BP model, and train the network until the mean square error of the network's performance index is less than the maximum permissible error e. max .

10. The method for assessing the residual life of stator bars in a hydro-generator unit based on adaptive operating conditions according to claim 1, characterized in that: Step S5 specifically includes the following process: In practical applications, based on the load curves issued by the dispatch center and the hydrological system data of the power station, the data is first divided into different operating period segments according to the set of operating conditions. The divided operating period segments are then input into the life loss benchmark model. Different benchmark models are adapted according to the conditions to predict the loss life under each operating condition. Finally, the overall remaining life of the bars is obtained by combining the historical life data.

Images