Pumped storage unit reliability grade prediction method and system

Abstract

The invention discloses a pumped storage group reliability grade prediction method and system, and relates to the technical field of power system optimization scheduling, and the method comprises the steps: obtaining an adjustment intensity index and a reliability score data set of a pumped storage group; carrying out reliability grade division on the pumped storage unit according to the reliability score; training the Adaboost integrated learning model by taking the adjustment intensity index as input and the reliability level of the pumped storage unit as output, and constructing an adjustment intensity-reliability level prediction model based on an Adaboost algorithm; and inputting the real-time adjustment intensity index of the pumped storage unit to be predicted into the adjustment intensity-reliability grade prediction model based on the Adaboost algorithm, and outputting to obtain the reliability grade of the pumped storage unit. According to the invention, operation reliability management and service life optimization of the pumped storage unit can be realized in a complex power grid environment with high-proportion new energy access.

Description

Technical Field

[0001] The present invention belongs to the technical field of optimal dispatching of power systems, and particularly relates to a method and system for predicting the reliability level of pumped-storage units. Background Art

[0002] With the continuous increase in the proportion of renewable energy generation, the dispatching complexity of power systems has also increased significantly. As a key energy storage technology, pumped-storage power stations have become an important means to balance the power supply and demand of power systems due to their good regulation performance. By reasonably dispatching pumped-storage units, the power supply and demand fluctuations brought about by the access of new energy can be effectively addressed, ensuring the safety and reliability of power grid operation. However, with the increase in the proportion of new energy in the power system, the operation reliability of pumped-storage units faces new challenges. The units need to frequently start, stop, and switch operating conditions to respond to new energy fluctuations, resulting in a significant increase in the operation intensity. At the same time, components of the units such as guide vanes, bearings, and bolts also face a higher risk of fatigue damage. These series of problems have a greater impact on the service life of the units and the economic benefits of the power stations. The current dispatching methods face many challenges in dealing with the operation reliability of pumped-storage power stations under the background of high proportion of new energy access. The main reason is that the reliability of pumped-storage power stations is affected by multiple factors, including the energy structure, the accuracy of new energy power prediction, the fluctuations of grid load demand, and the production quality and operation and maintenance management level of equipment. Therefore, how to predict the unit reliability in different dispatching scenarios has become an urgent problem to be solved.

[0003] Patent CN106447218B discloses a method for evaluating the reliability of a pumped-storage system based on multiple wind farms, including the following steps: S1: According to the correlation characteristics of the wind speeds of two adjacent wind farms, calculate and obtain the wind speed time series of the wind speeds of the two adjacent wind farms according to the ARMA method; S2: Establish a reliability evaluation model for the pumped-storage system of the wind farm; S3: Use the sequential Monte Carlo method to respectively evaluate the reliability of the reliability evaluation model of the pumped-storage system of the wind farm obtained in step S2. The beneficial effects of this invention are: closer to the actual wind speed correlation situation. During the evaluation process, the influences of the installed capacity, pumping and generating efficiency, and operation strategy of the pumped-storage power station are considered. The access of pumped-storage can reduce the amount of abandoned wind, improve the utilization rate of wind energy, and reduce the impact of wind power volatility on the safety of the power system.

[0004] The invention patent application CN115906429A discloses a reliability evaluation method, device, equipment and storage medium for a pumped-storage unit, which overall involves time-consuming costs and the effective number of supercapacitors for reliability. It selects not only the reliability effectiveness under a wide range of initial power states, but also uses a very low input power consumption time. The invention selects the highest and lowest points on the power increase regression line that can be used as reliability indicators under different initial power states, and can eliminate the influence of supercapacitor polarization on the evaluation of the reliability state of supercapacitors. The invention selects the reliability effectiveness suitable for different input and output power processes (that is, any increase in input and output power within the supercapacitor limit to the preset initial power state range), and obtains a more accurate reliability state evaluation deduction amount.

[0005] However, the prior art including the above-mentioned fails to cover the long-term impact of new energy fluctuations on the reliability of the unit and cannot achieve the prediction of reliability. Most of the prior art focuses on the economy of the wind-solar pumped-storage combined system and the reliability of the overall system, concentrating on the dispatching optimization and economic evaluation of the combined system, and less focusing on the reliability of the pumped-storage unit itself and its influencing mechanism. In addition, the existing unit reliability analysis also mostly relies on historical operation data, such as statistical models based on failure rates and equipment lifetimes, but most methods fail to fully consider the impact of new energy consumption requirements on the operation intensity of the unit. For example, although traditional methods can model the unit failure rate from the perspective of probability theory, they fail to deeply consider the fatigue damage and reliability impact of high-frequency start-stop and condition conversion on the pumped-storage unit, and it is difficult to effectively handle the dispatching pressure brought by large-scale new energy consumption. Summary of the Invention

[0006] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a method and system for predicting the reliability level of a pumped-storage unit. Based on adjusting the mapping relationship between the regulation intensity index and the reliability level of the pumped-storage unit, combined with the AdaBoost integrated learning algorithm, it automatically predicts the reliability level of the pumped-storage unit under different dispatching scenarios, and realizes the operation reliability management and life optimization of the pumped-storage unit in a complex power grid environment with high-proportion new energy access.

[0007] The present invention provides the following technical solutions: In the first aspect, a method for predicting the reliability level of a pumped-storage unit is provided, including: Obtain the regulation intensity index and the reliability score data set of the pumped-storage unit; Divide the reliability level of the pumped-storage unit according to the reliability score; Taking the regulation intensity index as the input and the reliability level of the pumped-storage unit as the output, train the Adaboost integrated learning model to construct a regulation intensity-reliability level prediction model based on the Adaboost algorithm; Input the real-time regulation intensity index of the pumped-storage unit to be predicted into the regulation intensity-reliability level prediction model based on the Adaboost algorithm, and output the reliability level of the pumped-storage unit; Among them, the method for constructing the regulation intensity-reliability level prediction model based on the Adaboost algorithm includes: Initialize the sample weights to make the weights of all input samples equal; Iteratively train the weak classifier. In each round of training, according to the weights of the current input samples, use the regulation intensity index as the input and the reliability level of the pumped-storage unit as the output to train the weak classifier; in each round of training, calculate the error rate of the current weak classifier, calculate the weight of the weak classifier according to the error rate of the weak classifier. At the same time, for the input samples with classification errors, increase their weights and update them as the new weights; After the iterative training is completed, weight the results of each round of weak classifiers according to their weights to form the final strong classifier, that is, the regulation intensity-reliability level prediction model based on the Adaboost algorithm.

[0008] Furthermore, the regulation intensity index includes the number of unit start-stop times, power generation operation hours, comprehensive utilization hours, regulation mileage, and vibration zone operation duration.

[0009] Furthermore, the reliability levels of the pumped-storage units include low, medium, good, and excellent, which are represented by labels 1, 2, 3, and 4 respectively.

[0010] Furthermore, after initializing the sample weights, the weight formula of the input sample is: ; In the formula, is the initial weight of the i th sample, N is the total number of samples.

[0011] Furthermore, the weak classifier adopts a decision tree; in each round of training, the calculation formula for the error rate of the current weak classifier is: ; In the formula, R t is the error rate of the weak classifier in the t th round of iterative training, is the weight of the t th sample in the i th round of iterative training,N is the total number of samples, x i is the i th sample, y i is the reliability level label of the pumped-storage unit corresponding to the i th sample; h t ( x i ) is the t th sample in the i th iteration; x i is the prediction result of the weak classifier for the is the indicator function. If h t ( x i ) is inconsistent with y i , it is 1; if h t ( x i ) is consistent with y i , it is 0.

[0012] Furthermore, the formula for calculating the weight of the weak classifier based on the error rate of the weak classifier is: ; In the formula, is the weight of the weak classifier in the t th iteration training, R t is the error rate of the weak classifier in the t th iteration training.

[0013] Furthermore, the formula for updating the weight of the misclassified input sample to a new weight after increasing its weight is: ; ; In the formula, is the weight of the t +1th iteration training for the i th sample, is the weight of the t th iteration training for the i th sample, is the weight of the weak classifier in the t th iteration training, y i is the reliability level label of the pumped-storage unit corresponding to the i th sample.h t ( x i ) is the t th sample in the i th iteration, x i and is the prediction result of the weak classifier for this sample. N represents the total number of samples.

[0014] Furthermore, the calculation formula of the strong classifier is: ; In the formula, H ( x ) is the strong classifier, sign (*) is the sign function, is the weight of the weak classifier in the t th iteration training, T is the total number of iterations, h t ( x ) is the weak classifier trained in the t th iteration, x is the input sample feature vector.

[0015] In a second aspect, a reliability level prediction system for a pumped-storage unit is provided, including: A data acquisition module for acquiring the regulation intensity index and the reliability score dataset of the pumped-storage unit; A reliability level division module for dividing the reliability level of the pumped-storage unit according to the reliability score; A model construction module for training an Adaboost ensemble learning model with the regulation intensity index as the input and the reliability level of the pumped-storage unit as the output, and constructing a regulation intensity-reliability level prediction model based on the Adaboost algorithm; A prediction module for inputting the real-time regulation intensity index of the pumped-storage unit to be predicted into the regulation intensity-reliability level prediction model based on the Adaboost algorithm, and outputting the reliability level of the pumped-storage unit; Among them, the method for constructing the regulation intensity-reliability level prediction model based on the Adaboost algorithm includes: Initializing the sample weights to make the weights of all input samples equal; Iteratively train the weak classifier. In each round of training, use the adjustment intensity index as the input and the reliability level of the pumped-storage unit as the output according to the weights of the current input samples to train the weak classifier; in each round of training, calculate the error rate of the current weak classifier, calculate the weight of the weak classifier according to the error rate of the weak classifier. At the same time, for the input samples with classification errors, increase their weights and update them as the new weights. After the iterative training is completed, weight the results of each round of weak classifiers according to their weights to form the final strong classifier, that is, the adjustment intensity-reliability level prediction model based on the Adaboost algorithm.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) Based on the mapping relationship between the adjustment intensity index and the reliability level of the pumped-storage unit, and combined with the AdaBoost ensemble learning algorithm, the present invention realizes the automatic prediction of the reliability level of the pumped-storage unit under different scheduling scenarios. On the one hand, through the analysis of the adjustment intensity index of the pumped-storage unit under the scheduling operation requirements, the prediction of the reliability level of the pumped-storage unit under different operating conditions is realized; on the other hand, the adjustment intensity-reliability level prediction model based on the Adaboost algorithm dynamically adjusts the classifier weights during the training process, enhancing the global learning ability of the model, and being able to efficiently predict the reliability level of the pumped-storage unit under future complex scheduling scenarios, providing a reliable decision-making basis for optimal scheduling and power station management, and endowing the method provided by the present invention with the ability of dynamic adjustment and global optimization; (2) The prediction method provided by the present invention can quickly adapt to scheduling requirements such as changes in the proportion of new energy access and load fluctuations. On the premise of meeting the power balance of the power system, through reliability assessment to support flexible scheduling, providing a strong guarantee for the stable and efficient operation of the pumped-storage unit, and realizing the operation reliability management and life optimization of the pumped-storage unit in a complex power grid environment with a high proportion of new energy access. Brief Description of the Drawings

[0017] Figure 1 is a schematic flow chart of the method for predicting the reliability level of the pumped-storage unit in an embodiment of the present invention; Figure 2 is a schematic construction flow chart of the adjustment intensity-reliability level prediction model based on the Adaboost algorithm in an embodiment of the present invention. Detailed Embodiments

[0018] The present invention will be further described below with reference to the drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and should not be used to limit the protection scope of the present invention.

[0019] The term "and / or" is merely a description of the relationship between associated objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone. Additionally, the character " / " generally indicates that the associated objects before and after are in an "or" relationship.

[0020] Embodiment 1

[0021] As Figure 1 shown, this embodiment provides a method for predicting the reliability level of a pumped-storage unit, including the following steps: Step 1: Obtain the regulation intensity index and reliability score dataset of the pumped-storage unit.

[0022] The regulation intensity index includes the number of unit starts and stops, power generation operation hours, comprehensive utilization hours, regulation mileage, and operation duration in the vibration area.

[0023] The reliability score is calculated based on the reliability comprehensive evaluation coefficient of the pumped-storage unit as stipulated in the "National Management Measures for Reliability Benchmarking of Generator Sets".

[0024] Step 2: Classify the reliability level of the pumped-storage unit according to the reliability score.

[0025] Since the numerical accuracy of the reliability score is too high, it is difficult to directly predict under limited data. To effectively extract the relationship between the regulation intensity and reliability of the pumped-storage unit, a theory for predicting the reliability level is proposed.

[0026] First, the reliability scores of the pumped-storage unit need to be divided into the required grade classifications. In this embodiment, it is set to four grades, including low, medium, good, and excellent, which are represented by labels 1, 2, 3, and 4 respectively.

[0027] Step 3: Use the regulation intensity index as the input and the reliability level of the pumped-storage unit as the output to train the Adaboost integrated learning model, and construct a regulation intensity-reliability level prediction model based on the Adaboost algorithm. As Figure 2 shown, the specific steps are as follows: Step 3.1: Initialize the weights.

[0028] To balance the importance of samples, the weights of all input samples are set to be equal in the initial stage, which is convenient for distinguishing samples with greater classification difficulty in the subsequent iterative process. The formula for initializing the weights of input samples is: ; In the formula, is the initial weight of the i th sample, NLet \(N\) be the total number of samples, and each sample has \(M\) features. In this embodiment, \(M = 5\), corresponding to 5 items of the adjustment intensity index.

[0029] Step 3.2: Iteratively train the weak classifier.

[0030] In each round of iterative training, according to the weights of the current input samples, using the adjustment intensity index as the input and the reliability level of the pumped-storage unit as the output, train the decision tree weak classifier to identify the relationship between each adjustment intensity index and the reliability level.

[0031] In each round of training, calculate the error rate of the current weak classifier, that is, the proportion of samples predicted incorrectly under the adjustment intensity index data. The higher the error rate, the worse the classification effect of the current model on this sample. Among them, the calculation formula for the error rate of the current weak classifier is: ; In the formula, R t is the error rate of the weak classifier in the t th round of iterative training, is the weight of the t rd sample in the i th round of iterative training, N is the total number of samples, x i is the i nd sample, y i is the reliability level label of the pumped-storage unit corresponding to the i th sample, h t ( x i ) is the prediction result of the weak classifier for the t th sample in the i th round of iteration; x i is the indicator function. If h ( t ) is not consistent with x i , it is 1; if y i ( h t ) is consistent with x i , it is 0. y i

[0032] Step 3.3: Adjust the weights of the weak classifier.

[0033] ​Calculate the weights of weak classifiers based on their error rates, such that weak classifiers with lower error rates have higher weights in the overall model, thus forming a strong classifier after multiple rounds of iteration. Among them, the calculation formula for the weights of weak classifiers is: ; In the formula, is the weight of the weak classifier in the t -th round of iterative training, R t is the error rate of the weak classifier in the t -th round of iterative training.

[0034] Step 3.4: Update the sample weights.

[0035] For the input samples that are misclassified, after increasing their weights, update them to new weights so that the weak classifier can learn these samples more accurately in the next round of iteration. This process continues iteratively, finally focusing on the samples that are difficult to distinguish, in order to improve the overall classification effect of the model. Among them, the calculation formula for the new updated weights is: ; ; In the formula, is the weight of the t +1-th sample in the i -th round of iterative training, is the weight of the t -th sample in the i -th round of iterative training, is the weight of the weak classifier in the t -th round of iterative training, y i is the reliability level label of the pumped-storage unit corresponding to the i -th sample, h t ( x i ) is the prediction result of the weak classifier for the t -th sample in the i -th round of iteration x i , N is the total number of samples.

[0036] Step 3.4: Construct a strong classifier model.

[0037] After the iterative training is completed, weight the results of each round of weak classifiers according to their weights to form the final strong classifier, that is, the regulation strength-reliability level prediction model based on the Adaboost algorithm. Among them, the calculation formula for the strong classifier is: ; Wherein, H ( x ) is a strong classifier, sign (*) is a sign function that returns 1 if * is greater than 0 and -1 otherwise; is the weight of the weak classifier in the t th round of iterative training, T is the total number of iterations, h t ( x ) is the weak classifier trained in the t th round of iteration, x is the input sample feature vector.

[0038] Step 4: Input the real-time regulation intensity index of the pumped-storage unit to be predicted into the regulation intensity-reliability level prediction model based on the Adaboost algorithm, and output the reliability level of the pumped-storage unit.

[0039] Example 2

[0040] In this example, the method of Example 1 is used to predict the reliability level of the pumped-storage unit in combination with an example.

[0041] In this example, 40 groups of regulation intensity index and reliability score data are taken, and part of the data is shown in Table 1.

[0042] Table 1 Sample data of regulation intensity index and reliability score of pumped-storage units

[0043] Based on the historical reliability scores of the units in Table 1, the reliability is divided into certain high and low levels, and classification operations are performed on it. In this example, it is set to four levels, namely low, medium, good, and excellent (corresponding to classification numbers 1 to 4 respectively), as shown in Table 2.

[0044] Table 2 Reliability level division of pumped-storage units

[0045] The sample data after dividing the levels is shown in Table 3.

[0046] Table 3 Sample data after dividing the levels

[0047] Set the ratio of the training set to the test set of the above 40 sample data to 3:1, that is, randomly select 30 samples for the training set and 10 samples for the test set. It is stipulated that the number of decision tree weak classifiers is 30, and the training and prediction results are shown in Table 4.

[0048] Table 4 Prediction results

[0049] As can be seen from Table 4, tests on the reliability level prediction of pumped-storage units were carried out on 10 sample data. The results show that only one result is incorrect, and the remaining test samples can all achieve the prediction of the unit operation reliability level through the regulation intensity index data of the pumped-storage unit. That is, the accuracy rate of the reliability classification prediction using the method provided by the present invention is 90%. If there is more abundant historical data as support in the future, the learning prediction analysis accuracy can be further improved, which can be used to support the refined research on the variation law of the unit reliability under different operation scenarios when operating in combination with large-scale new energy.

[0050] Embodiment 3

[0051] Based on the same inventive concept as Embodiment 1, this embodiment provides a reliability level prediction system for pumped-storage units, including: A data acquisition module, configured to acquire the regulation intensity index and the reliability score data set of the pumped-storage unit; A reliability level division module, configured to divide the reliability level of the pumped-storage unit according to the reliability score; A model construction module, configured to train the Adaboost ensemble learning model with the regulation intensity index as the input and the reliability level of the pumped-storage unit as the output, and construct a regulation intensity-reliability level prediction model based on the Adaboost algorithm; A prediction module, configured to input the real-time regulation intensity index of the pumped-storage unit to be predicted into the regulation intensity-reliability level prediction model based on the Adaboost algorithm, and output the reliability level of the pumped-storage unit.

[0052] For the specific function implementation of the above modules, refer to the relevant content in the method of Embodiment 1, which will not be elaborated here.

[0053] Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0054] This application is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing device produce means for implementing the functions specified in one or more of the flows Figure 1 one or more of the flows and / or blocks Figure 1 or means for implementing the functions specified in one or more of the blocks.

[0055] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including instruction means for implementing the functions specified in one or more of the flows Figure 1 one or more of the flows and / or blocks Figure 1 or means for implementing the functions specified in one or more of the blocks.

[0056] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operational steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more of the flows Figure 1 one or more of the flows and / or blocks Figure 1 or means for implementing the functions specified in one or more of the blocks.

[0057] The above is only the preferred embodiment of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the technical principle of the present invention, several improvements and modifications can be made, and these improvements and modifications should also be regarded as the protection scope of the present invention.

Claims

1. A method for predicting the reliability level of a pumped-storage unit, characterized in that, including: obtaining a dataset of the regulation intensity index and reliability scores of a pumped-storage unit; classifying the reliability levels of the pumped-storage unit according to the reliability scores; using the regulation intensity index as the input and the reliability level of the pumped-storage unit as the output to train an Adaboost ensemble learning model, and constructing a regulation intensity-reliability level prediction model based on the Adaboost algorithm; inputting the real-time regulation intensity index of the pumped-storage unit to be predicted into the regulation intensity-reliability level prediction model based on the Adaboost algorithm, and outputting the reliability level of the pumped-storage unit; wherein the method for constructing the regulation intensity-reliability level prediction model based on the Adaboost algorithm includes: initializing the sample weights to make the weights of all input samples equal; iteratively training the weak classifier. In each round of training, according to the weights of the current input samples, using the regulation intensity index as the input and the reliability level of the pumped-storage unit as the output to train the weak classifier; in each round of training, calculating the error rate of the current weak classifier, calculating the weight of the weak classifier according to the error rate of the weak classifier, and at the same time, for the input samples with classification errors, increasing their weights and then updating them as the new weights; after the iterative training is completed, weighting the results of each round of weak classifiers according to their weights to form the final strong classifier, that is, the regulation intensity-reliability level prediction model based on the Adaboost algorithm.

2. The reliability level prediction method of the pumped storage unit according to claim 1, characterized in that The regulation intensity index includes the number of unit starts and stops, the power generation operation hours, the comprehensive utilization hours, the regulation mileage, and the operation duration in the vibration area.

3. The reliability level prediction method for a pumped storage unit according to claim 1, characterized in that, The reliability levels of the pumped-storage unit include low, medium, good, and excellent, which are represented by labels 1, 2, 3, and 4 respectively.

4. The reliability level prediction method for a pumped storage unit according to claim 1, characterized in that After the sample weights are initialized, the weight formula of the input samples is: ； Wherein, is the initial weight of the i th sample, N is the total number of samples.

5. The reliability level prediction method of the pumped storage unit according to claim 1, characterized in that The weak classifier uses a decision tree; in each round of training, the calculation formula for the error rate of the current weak classifier is: ； In the formula, R t For the t The error rate of the weak classifier in the round of iterative training, For the t In the round of iterative training i The weight of the samples, N is the total number of samples, x i For the i samples, y i For the i The reliability level label of the pumped storage unit corresponding to the sample, h t ( x i ) is the t In the iteration i Samples x i The weak classifier prediction results; is an indicator function, if h t ( x i )and y i If they are not consistent, then it is 1; if h t ( x i )and y i If they are consistent, it is 0.

6. The reliability level prediction method of the pumped storage unit according to claim 1, wherein The formula for calculating the weight of the weak classifier according to the error rate of the weak classifier is: ； In the formula, is the weight of the weak classifier in the t -th round of iterative training, R t is the error rate of the weak classifier in the t -th round of iterative training.

7. The reliability level prediction method for a pumped storage unit according to claim 1, characterized in that For the input samples with classification errors, the formula for increasing their weights and then updating them as the new weights is: ； ； In the formula, is the weight of the t -th sample in the i +1-th round of iterative training, is the weight of the t -th sample in the i -th round of iterative training, is the weight of the weak classifier in the t -th round of iterative training, y i is the reliability level label of the pumped-storage unit corresponding to the i -th sample, h t ( x i ) is the prediction result of the weak classifier for the t -th sample in the i -th round of iteration, x i , N is the total number of samples.

8. The reliability level prediction method for a pumped-storage unit according to claim 1, wherein The calculation formula for the strong classifier is: ； In the formula, H ( x ) is a strong classifier, sign (*) is the sign function, is the weight of the weak classifier in the t th round of iterative training, T is the total number of iterations, h t ( x ) is the weak classifier trained in the t th round of iteration, x is the input sample feature vector.

9. A reliability level prediction system for a pumped-storage unit, characterized in that, including: a data acquisition module for obtaining a dataset of the regulation intensity index and reliability scores of a pumped-storage unit; a reliability level classification module for classifying the reliability levels of the pumped-storage unit according to the reliability scores; a model construction module for using the regulation intensity index as the input and the reliability level of the pumped-storage unit as the output to train an Adaboost ensemble learning model, and constructing a regulation intensity-reliability level prediction model based on the Adaboost algorithm; a prediction module for inputting the real-time regulation intensity index of the pumped-storage unit to be predicted into the regulation intensity-reliability level prediction model based on the Adaboost algorithm, and outputting the reliability level of the pumped-storage unit; wherein the method for constructing the regulation intensity-reliability level prediction model based on the Adaboost algorithm includes: initializing the sample weights to make the weights of all input samples equal; Iteratively train the weak classifier. In each round of training, use the adjustment intensity index as the input and the reliability level of the pumped-storage unit as the output according to the weights of the current input samples to train the weak classifier; in each round of training, calculate the error rate of the current weak classifier, calculate the weight of the weak classifier according to the error rate of the weak classifier. At the same time, for the input samples with classification errors, increase their weights and update them as the new weights. After the iterative training is completed, weight the results of each round of weak classifiers according to their weights to form the final strong classifier, that is, the adjustment intensity-reliability level prediction model based on the Adaboost algorithm.

Images