Medium and long term runoff integrated prediction method based on multi-evaluation index integration

Abstract

According to the medium and long term runoff integrated prediction method based on multi-evaluation index integration provided by the invention, the performance of the prediction model is evaluated by comprehensively considering a plurality of evaluation indexes, the expert experience and the machine learning method are integrated to distribute the index weight, and the accuracy and the stability of the prediction model are improved. The method is suitable for the field of runoff prediction, and provides a more reliable basis for runoff prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of runoff prediction, and in particular to a medium and long-term runoff integrated prediction method based on comprehensive multi-evaluation indexes. Background Art

[0002] In the existing prediction technologies, a single prediction model often fails to comprehensively consider the complexity and diversity of data, resulting in deviation in the prediction results. To solve this problem, a dynamic integrated prediction method has emerged, which improves the prediction accuracy by integrating the results of multiple prediction models. However, how to reasonably allocate the weights of each model is a key issue in the dynamic integrated prediction method. Traditional weight allocation methods often only consider a single index such as prediction accuracy and ignore the performance of the model in other aspects. Summary of the Invention

[0003] The purpose of the present invention is to provide a medium and long-term runoff integrated prediction method based on comprehensive multi-evaluation indexes for the deficiencies of the above-mentioned existing technologies, aiming to comprehensively consider the performance of the prediction model in different aspects, so as to allocate weights more scientifically and reasonably and improve the prediction accuracy and stability.

[0004] To achieve the above purpose, the present invention adopts the following technical solutions:

[0005] The present invention provides a medium and long-term runoff integrated prediction method based on comprehensive multi-evaluation indexes, including the following steps:

[0006] S1. Construction of evaluation index matrix: Select multiple evaluation indexes, construct the corresponding evaluation index matrix, and perform standardization processing on the corresponding evaluation index matrix to obtain the standardized evaluation index matrix;

[0007] S2. Determination of evaluation index weights: Determine the final weight vector of each evaluation index by a subjective and objective combined weighting method;

[0008] S3. Calculation of comprehensive score: Calculate the comprehensive score of each model according to the standardized evaluation index matrix and the final weight vector of the evaluation index;

[0009] S4. Weight allocation: Allocate the final model weights by a normalization method according to the comprehensive scores of each model;

[0010] S5. Dynamic integrated prediction: Perform weighted averaging on the results of multiple prediction models using the allocated model weights to obtain the final prediction result.

[0011] Further, the specific steps of S1 are as follows:

[0012] S101. Select multiple evaluation metrics to measure the performance of the prediction model in different aspects. Each evaluation metric can reflect the model from a certain perspective;

[0013] S102. Construct an evaluation metric matrix: Let A be an n×m evaluation metric matrix, and the evaluation metric matrix A is:

[0014]

[0015] where, a nm is the evaluation value of the nth model on the mth metric;

[0016] S103. Standardize the evaluation metrics: Perform standardization processing on the evaluation metric matrix A to obtain the standardized evaluation metric matrix A′:

[0017]

[0018] where, a′ nm is the standardized evaluation value of the nth model on the mth metric;

[0019] a′ ij is standardized by the min-max method, specifically:

[0020]

[0021] where, a′ ij is the standardized performance of the ith model on the jth evaluation metric.

[0022] Furthermore, the specific content of S2 is as follows:

[0023] S201. Rank the importance among the metrics to form an evaluation metric importance ranking matrix. The evaluation metric importance ranking matrix is represented as an m-dimensional vector, denoted as R = (r1, r2, r j …r m ), where, r j is the importance ranking of the jth metric;

[0024] S202. According to the importance ranking, set the weight coefficients between different metrics in a randomly decreasing manner;

[0025] Set a base weight value w base and a decreasing factor δ to control the speed of weight decrease with the ranking. The value range of the decreasing factor δ is between 0 and 1. The closer it is to 1, the faster the weight decreases. For each metric, introduce a random perturbation term ε j , then the weight w j ' 指标 of the jth metric is expressed as:

[0026]

[0027] Among them, is the weight part according to the decreasing ranking. The higher the ranking, the greater the weight; (1 + ε j ) is the random perturbation term, which is used to increase the randomness of the weight;

[0028] Thus, a preliminary weight coefficient vector

[0029] S203. After obtaining the preliminary weight coefficient vector W' 指标 further adjust and optimize the weight through historical data;

[0030] Thus, a training set containing historical indicators and optimal model data is established, and a supervised machine learning algorithm neural network is used to train the model. During the training process, the preliminary weight coefficient vector W' 指标 is used as the initial weight of the model, and the weight is continuously adjusted by the gradient descent method to minimize the prediction error of the model on the training set;

[0031] After training, the final weight vector is obtained, where w j 指标 is the weight of the jth indicator and satisfies the condition.

[0032] Furthermore, the specific content of S3 is as follows:

[0033] According to the standardized evaluation index matrix A′ and the evaluation index weight w 指标 , calculate the comprehensive score Score i of each model. The calculation formula of the comprehensive score is:

[0034]

[0035] where Score i represents the comprehensive score of the ith model.

[0036] Furthermore, the specific content of S4 is as follows:

[0037] Allocate the final weight according to the comprehensive score of each model Adopt the normalization method, and the calculation formula of the weight is:

[0038]

[0039] where is the final weight of the ith model.

[0040] Further, S5 is specifically as follows:

[0041] Use the allocated final weights to perform weighted averaging on the results of multiple prediction models to obtain the final prediction result:

[0042]

[0043] where model i is the prediction result of the i-th model.

[0044] The beneficial effects of the present invention are as follows: By comprehensively considering multiple evaluation indicators, the performance of the prediction model is evaluated more comprehensively, improving the scientificity and rationality of weight allocation. Using the standardization process and normalization method, the dimensional differences between different evaluation indicators are eliminated, making the weight allocation more accurate. The accuracy and stability of the prediction model are improved, providing a more reliable basis for data analysis and decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is a flowchart of a medium and long-term runoff integrated prediction method based on multi-evaluation index synthesis;

[0046] Figure 2 is a schematic diagram of the prediction effect of the runoff collection of the Three Gorges Reservoir from January to December. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0047] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0048] A medium and long-term runoff integrated prediction method based on multi-evaluation index synthesis includes the following steps:

[0049] S1. Construction of evaluation index matrix: Select multiple evaluation indicators, construct the corresponding evaluation index matrix, and perform standardization processing on the corresponding evaluation index matrix to obtain the standardized evaluation index matrix;

[0050] S2. Determination of evaluation index weights: Determine the final weight vector of each evaluation indicator by a method of combining subjective and objective weight assignment;

[0051] S3. Calculation of comprehensive score: Calculate the comprehensive score of each model according to the standardized evaluation index matrix and the final weight vector of the evaluation indicators;

[0052] S4. Weight allocation: According to the comprehensive scores of each model, use the normalization method to allocate the final model weights;

[0053] S5, Dynamic Integrated Prediction: Use the assigned model weights to perform weighted averaging on the results of multiple prediction models to obtain the final prediction result.

[0054] The specific steps of S1 are as follows:

[0055] S101, Select multiple evaluation metrics to measure the performance of the prediction model in different aspects. Each evaluation metric can reflect the model from a certain perspective;

[0056] In this embodiment, prediction accuracy, stability, and computational efficiency are selected as evaluation metrics.

[0057] S102, Construct an evaluation metric matrix: In this embodiment, there are two prediction models, namely a multiple regression model and a random forest model. The evaluation metric matrix A is:

[0058]

[0059] S103, Standardize the evaluation metrics: Use the min-max method to standardize the evaluation metric matrix A to obtain the standardized evaluation metric matrix A';

[0060]

[0061] The specific content of S2 is as follows:

[0062] S201, Rank the importance among the metrics to form an evaluation metric importance ranking matrix. The evaluation metric importance ranking matrix is represented as a 3D vector, where each element of the vector represents the importance ranking of a metric, denoted as R = (1, 2, 3), representing the importance rankings of prediction accuracy, stability, and computational efficiency respectively.

[0063] S202, According to the importance ranking, set the weight coefficients between different metrics in a randomly decreasing manner to ensure the weight distribution can reflect the importance of the metrics and have a certain degree of flexibility.

[0064] Set a base weight value w base and a decreasing factor δ, which is used to control the speed at which the weight decreases with the ranking. The value range of the decreasing factor δ is between 0 and 1. The closer it is to 1, the faster the weight decreases. For each metric, introduce a random perturbation term ε j , which is used to increase the flexibility of the weight. ε j is a random number within a small range to ensure that the weight distribution does not deviate too much from the base distribution based on the ranking.

[0065] Then the weight w j ' 指标 of the j-th metric is expressed as:

[0066]

[0067] Among them, is the weight part according to the decreasing ranking. The higher the ranking, the greater the weight; (1 + ε j ) is the random perturbation term, which is used to increase the randomness of the weight;

[0068] Thus, a preliminary weight coefficient vector

[0069] S203. After obtaining the preliminary weight coefficient vector W' 指标 , further adjust and optimize the weight through historical data;

[0070] Thus, a training set containing historical indicators and optimal model data is established,

[0071] D = {(x1, y1), (x2, y2),..., (x k , y k )..., (x n , y n )}, where x k is the feature vector of the kth sample; y k is the optimal model corresponding to the kth sample;

[0072] Adopt the supervised machine learning algorithm neural network to train the model. During the training process, use the preliminary weight coefficient vector W' 指标 as the initial weight of the model, and continuously adjust the weight through the gradient descent method to minimize the prediction error of the model on the training set;

[0073] After training, obtain the final weight vector where w j index is the weight of the jth index and satisfies the condition.

[0074] In this embodiment, the finally determined weights for prediction accuracy, stability, and computational efficiency are 0.5, 0.3, and 0.2 respectively.

[0075] Specifically, S3 is as follows:

[0076] According to the standardized evaluation index matrix A' and the evaluation index weight w 指标 , calculate the comprehensive score Score i of each model. The calculation formula for the comprehensive score is:

[0077]

[0078] where Score i represents the comprehensive score of the ith model.

[0079] In this embodiment, the comprehensive scores Score1 and Score2 of the two prediction models are calculated as follows:

[0080] Score1 = 0.5×1.00 + 0.3×0.00 + 0.2×0.00 = 0.50;

[0081] Score2 = 0.5×0.00 + 0.3×1.00 + 0.2×1.00 = 0.50;

[0082] Here, due to the existence of the same maximum and minimum values in the standardized evaluation index matrix A′, the scores of the two models on some evaluation indexes are the same, resulting in the same comprehensive scores;

[0083] Specifically, S4 is as follows:

[0084] Allocate the final weights according to the comprehensive scores of each model Using the normalization method, the calculation formula of the weight is:

[0085]

[0086] where, is the final weight of the i-th model. Due to the same scores, the final weights of the two models in this embodiment are both 0.5.

[0087] Specifically, S5 is as follows:

[0088] Use the allocated final weights Perform weighted averaging on the results of multiple prediction models to obtain the final prediction result:

[0089]

[0090] where, model i is the prediction result of the i-th model.

[0091] In the embodiment, the final prediction result is the average result of the two models, as shown in Figure 2 .

[0092] In summary, the present invention provides a medium and long-term runoff integrated prediction method and system based on comprehensive multi-evaluation indexes. By comprehensively considering multiple evaluation indexes, the performance of the prediction model is evaluated more comprehensively, and the scientificity and rationality of weight allocation are improved. The standardization processing and normalization method are adopted to eliminate the dimensional difference between different evaluation indexes, making the weight allocation more accurate, improving the accuracy and stability of the prediction model, and facilitating the application and popularization of medium and long-term runoff prediction services.

[0093] The embodiments described above only represent the implementation manners of the present invention. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the patent for the present invention. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all fall within the protection scope of the present invention. Therefore, the protection scope of the patent for the present invention shall be subject to the appended claims.

Claims

1. A medium- and long-term runoff integrated prediction method based on comprehensive multi-evaluation indicators, characterized in that: The following steps are involved: S1. Construction of evaluation indicator matrix: Select multiple evaluation indicators, construct a corresponding evaluation indicator matrix, and standardize the corresponding evaluation indicator matrix to obtain a standardized evaluation indicator matrix; S2. Determine the weight of evaluation indicators: Determine the final weight vector of each evaluation indicator through the subjective and objective combined weighting method; S3. Calculate the comprehensive score: Calculate the comprehensive score of each model based on the standardized evaluation indicator matrix and the final weight vector of the evaluation indicators; S4. Assign weights: According to the comprehensive score of each model, the final model weight is assigned using the normalization method; S5. Dynamic integrated prediction: Use the assigned model weights to perform weighted averaging on the results of multiple prediction models to obtain the final prediction result.

2. The method for mid- to long-term runoff integrated prediction based on comprehensive multi-evaluation indicators according to claim 1 is characterized in that: The specific steps of S1 are: S101. Select multiple evaluation indicators to measure the performance of the prediction model in different aspects. Each evaluation indicator can reflect the model from one perspective. S102, constructing an evaluation index matrix: Let A be an n×m evaluation index matrix, and the evaluation index matrix A is: Among them, a nm is the evaluation value of the nth model on the mth indicator; S103, standardizing evaluation indicators: standardizing the evaluation indicator matrix A to obtain a standardized evaluation indicator matrix A′: Among them, a n ' m is the standardized evaluation value of the nth model on the mth indicator; The value of ai′j is standardized using the minimum-maximum method, specifically: Among them, ai′j is the performance of the i-th model after normalization on the j-th evaluation indicator.

3. The method for mid- to long-term runoff integrated prediction based on comprehensive multi-evaluation indicators according to claim 2 is characterized in that: The S2 is specifically: S201, sort the importance of the indicators to form an evaluation indicator importance sorting matrix, the evaluation indicator importance sorting matrix is ​​represented as an m-dimensional vector, denoted as R = (r1, r2, r j …r m ), where rj is the importance ranking of the jth indicator; S202, according to the importance ranking, the weight coefficients between different indicators are set in a random decreasing manner; Set a basic weight value wbase and a decreasing factor δ to control the speed at which the weight decreases with the ranking. The value range of the decreasing factor δ is between 0 and 1. The closer it is to 1, the faster the weight decreases. For each indicator, introduce a random disturbance term ε j , then the weight w of the jth indicator j ' 指标 The indicator is expressed as: in, It is the weight part that decreases according to the ranking. The higher the ranking, the greater the weight. (1+εj) is a random perturbation term, which is used to increase the randomness of the weight. Thus, we get a preliminary weight coefficient vector S203, after obtaining the preliminary weight coefficient vector W' 指标 Finally, the weights are further adjusted and optimized through historical data; Thus, a training set containing historical indicators and optimal model data is established, and a supervised machine learning algorithm neural network is used to train the model. During the training process, the initial weight coefficient vector W' 指标 As the initial weight of the model, the weight is continuously adjusted through the gradient descent method to minimize the prediction error of the model on the training set; After training, the final weight vector is obtained Among them, the wj index is j The weight of the indicators, and meet conditions.

4. The method for mid- to long-term runoff integrated prediction based on comprehensive multi-evaluation indicators according to claim 3 is characterized in that: The S3 is specifically: According to the standardized evaluation index matrix A′ and the evaluation index weight w index, the comprehensive score Scorei of each model is calculated. The calculation formula of the comprehensive score is: Among them, Scorei represents the comprehensive score of the i-th model.

5. The method for mid- to long-term runoff integrated prediction based on comprehensive multi-evaluation indexes according to claim 4 is characterized in that: The S4 is specifically: The final weight is assigned according to the comprehensive score of each model Using the normalization method, the weight calculation formula is: in, is the final weight of the ith model.

6. The method for mid- to long-term runoff integrated prediction based on comprehensive multi-evaluation indexes according to claim 5 is characterized in that: The S5 is specifically: Using the final weight assigned The results of multiple prediction models are weighted averaged to obtain the final prediction result: Among them, model i is the prediction result of the ith model.

Images