Canal upstream water level prediction method based on data mechanism fusion model

Abstract

The invention discloses a method for predicting the upstream water level of a channel pool based on a data mechanism fusion model. The method comprises the following steps of: 1, constructing a similar prediction historical sequence set by utilizing a historical data set; 2, constructing an integral time-delay model to predict the downstream water level of the canal pool in a future time period; 3, taking a channel pool downstream water level sequence result obtained by the integral time delay prediction model as a correlation downstream water level sequence of a similarity analysis model, constructing a similarity prediction model through a Manhattan distance, predicting a channel pool downstream water level which is most similar in history, determining a history same-period scene, and predicting a channel pool upstream water level in the history same-period scene; according to the method, the uncertainty of correlation sequence deviation of a similar prediction model is overcome, and the prediction precision of the upstream water level of the canal pool is remarkably improved.

Description

Technical Field

[0001] The invention relates to the technical field of hydrological prediction, and in particular to a method for predicting water level in the upstream of a canal or pond based on a data mechanism fusion model. Background Art

[0002] The water level prediction of the canal pool is related to the water delivery scheduling safety and water supply efficiency of the canal. The high-precision water level prediction model can assist the scheduling decision-makers to formulate favorable measures in advance, avoid water delivery safety hazards and improve water supply efficiency. In recent years, a large number of mechanism-driven prediction models and data-driven prediction models have been applied to canal water level prediction. Among them, the mechanism-driven prediction model represented by the one-dimensional hydrodynamic model is widely used in the prediction of canal water surface line, but the construction process is complicated and the accuracy is subject to parameter calibration. The data-driven model represented by deep learning is also gradually used for canal water level prediction. The prediction speed is fast, but it lacks interpretability and the model training time is usually long. These two types of models usually appear in the form of a single model. There are also studies that use the coupling model of the two. Although it further improves the accuracy of water level prediction and improves the defect of the uninterpretability of the data-driven model, the model construction and training process is more complicated and often occupies a lot of computing resources. Summary of the invention

[0003] The purpose of the present invention is to overcome the above-mentioned shortcomings and provide a method for predicting the upstream water level of a canal based on a data mechanism fusion model. In view of the shortcomings of the existing canal water level prediction model, a computationally efficient integral time-lag model is considered as a mechanism-driven model to predict the downstream water level of the canal and use it as a related downstream water level series for similar predictions. A time series similarity analysis model is used to further predict the upstream water level series, thereby improving the interpretability and prediction accuracy of the prediction model.

[0004] In order to solve the above technical problems, the technical solution adopted by the present invention is: a method for predicting the upstream water level of a canal pool based on a data mechanism fusion model, which comprises the following steps:

[0005] Step 1: Use historical data sets to construct similar prediction historical sequence groups;

[0006] Step 2: Construct an integrated time-lag model to predict the downstream water level of the canal pond in the future period;

[0007] Step 3: The downstream water level sequence of the canal pool obtained by the integrated time-lag prediction model is used as the correlation downstream water level sequence of the similarity analysis model. A similarity prediction model is constructed through Manhattan distance to predict the most similar downstream water level of the canal pool, determine the historical scenario at the same time, and predict the upstream water level of the canal pool under the same historical scenario.

[0008] Furthermore, the step 1 specifically includes the following contents:

[0009] First, the historical river and canal water delivery data is cleaned, including the upstream flow, downstream flow, diversion flow, and downstream water level data of the river and canal, to ensure the integrity and accuracy of the data at each time point, and obtain the upstream flow, downstream flow, diversion flow, and downstream water level sequence set with a total time length of L;

[0010] Define the relevant sequence time length of the historical similar scenes of the similar prediction model as p, select the upstream water level and downstream water level long sequence historical data of the canal as the similar prediction historical data set, and divide the upstream water level and downstream water level long sequence historical data into N similar prediction historical sequence group sets according to the length of p, where each sequence group contains the upstream water level sequence and downstream water level sequence of length p;

[0011] The total number of similar prediction sequence groups is calculated as follows:

[0012] N = L-p+1;

[0013] Where L is the total time length of the historical data set.

[0014] Furthermore, the step 2 specifically includes the following contents:

[0015] Taking p+1 upstream flows, downstream flows, diversion flows and initial downstream water levels in the future period as input, the next P downstream water levels are predicted to obtain a downstream water level sequence of the channel pool with a length of p. The specific prediction formula is as follows:

[0016]

[0017] Where s(t) is the water depth H downstream of the canal at time t. d (t) Relative to the downstream water depth H at the initial time d (0), q in (t-τ) is the inflow flow rate Q of the canal pool at time (t-τ) in (t-τ) relative to the inflow flow q at the initial time in (0), q out (t) represents the outflow rate Q of the canal pool at time t out (t) relative to the outflow flow rate Q at the initial moment out (0), d(t) represents the increment of the diversion outlet flow rate D(t) at time t relative to the diversion outlet flow rate D(0) at the initial time, and the parameters of the backwater area A and the hysteresis time τ are obtained by hydrodynamic model identification.

[0018] Furthermore, the step 3 specifically includes the following contents:

[0019] The downstream water level sequence predicted by the integrated time-lag model is used as the correlation downstream water level sequence data of the similarity analysis model. The Manhattan distance between the correlation downstream water level sequence and each group of downstream water level sequences in the similar prediction historical sequence group is calculated. The historical downstream water level sequence with the smallest Manhattan distance is taken as the most similar historical downstream water level sequence. The Manhattan distance calculation formula is as follows:

[0020]

[0021] In the formula, l d,i is the downstream water level at the i-th time node in the integral time-delay result sequence, l' d,i is the downstream water level at the i-th time node in any sequence group in the historical sequence group set;

[0022] The most similar upstream water level sequence in history is calculated through the most similar downstream water level sequence in history. The calculation formula is as follows:

[0023]

[0024] In the formula, is the downstream water level of the i-th time node of the most similar downstream water level sequence in the historical sequence group set, is the upstream water level at the i-th time node of the most similar upstream water level sequence in the historical sequence group set;

[0025] The upstream water level of the canal is predicted by the most similar upstream water level in history. The calculation formula is as follows:

[0026]

[0027] In the formula, l u,i is the prediction result of the upstream water level of the canal, and β is the prediction correction coefficient of the upstream water level.

[0028] Furthermore, the calculation formula of the prediction correction coefficient of the upstream water level is as follows:

[0029]

[0030] Beneficial effects of the invention: In view of the shortcomings of the existing river and canal water level prediction models, the present invention considers adopting a computationally efficient integral time-lag model as a mechanism-driven model to predict the downstream water level of the river and canal and use it as a correlated downstream water level series for similar predictions, and adopts a time series similarity analysis model to further predict the upstream water level series. It fully utilizes the advantages of the predictability of the integral time-lag model and the efficiency of the similar prediction model, overcomes the uncertainty of the correlated sequence deviation of the similar prediction model, and significantly improves the prediction accuracy of the upstream water level of the canal. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1It is a flow chart of a method for predicting the upstream water level of a canal pool based on a data mechanism fusion model;

[0032] Figure 2 This is a comparison chart of the RMSE results of the method proposed in the present invention and similar prediction methods;

[0033] Figure 3 This is a comparison chart of the MAPE results of the method proposed in the present invention and similar prediction methods;

[0034] Figure 4 This is a comparison chart of the MSE results of the method proposed in the present invention and similar prediction methods;

[0035] Figure 5 This is a comparison chart of the MAE results between the method proposed in the present invention and similar prediction methods. DETAILED DESCRIPTION

[0036] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0037] Example 1: Figure 1 As shown, a method for predicting the upstream water level of a canal pool based on a data mechanism fusion model includes the following steps:

[0038] Step 1: Use historical data sets to construct similar prediction historical sequence groups;

[0039] First, the historical river and canal water delivery data is cleaned, including the upstream flow, downstream flow, diversion flow, and downstream water level data of the river and canal, to ensure the integrity and accuracy of the data at each time point, and obtain the upstream flow, downstream flow, diversion flow, and downstream water level sequence set with a total time length of L;

[0040] Define the relevant sequence time length of the historical similar scenes of the similar prediction model as p, select the upstream water level and downstream water level long sequence historical data of the canal as the similar prediction historical data set, and divide the upstream water level and downstream water level long sequence historical data into N similar prediction historical sequence group sets according to the length of p, where each sequence group contains the upstream water level sequence and downstream water level sequence of length p;

[0041] The total number of similar prediction sequence groups is calculated as follows:

[0042] N = L-p+1;

[0043] Where L is the total time length of the historical data set.

[0044] Step 2: Construct an integrated time-lag model to predict the downstream water level of the canal pond in the future period;

[0045] Taking p+1 upstream flows, downstream flows, diversion flows and initial downstream water levels in the future period as input, the next P downstream water levels are predicted to obtain a downstream water level sequence of the channel pool with a length of p. The specific prediction formula is as follows:

[0046]

[0047] Where s(t) is the water depth H downstream of the canal at time t. d (t) relative to the downstream water depth H at the initial time d (0), q in (t-τ) is the inflow flow rate Q of the canal pool at time (t-τ) in (t-τ) relative to the inflow flow q at the initial time in (0), q out (t) represents the outflow rate Q of the canal pool at time t out (t) relative to the outflow flow rate Q at the initial moment out (0), d(t) represents the increment of the diversion outlet flow rate D(t) at time t relative to the diversion outlet flow rate D(0) at the initial time, and the parameters of the backwater area A and the hysteresis time τ are obtained by hydrodynamic model identification.

[0048] Step 3: The downstream water level sequence of the canal pool obtained by the integrated time-lag prediction model is used as the correlation downstream water level sequence of the similarity analysis model. A similarity prediction model is constructed through Manhattan distance to predict the most similar downstream water level of the canal pool, determine the historical scenario at the same time, and predict the upstream water level of the canal pool under the same historical scenario.

[0049] The downstream water level sequence predicted by the integrated time-lag model is used as the correlation downstream water level sequence data of the similarity analysis model. The Manhattan distance between the correlation downstream water level sequence and each group of downstream water level sequences in the similar prediction historical sequence group is calculated. The historical downstream water level sequence with the smallest Manhattan distance is taken as the most similar historical downstream water level sequence. The Manhattan distance calculation formula is as follows:

[0050]

[0051] In the formula, l d,i is the downstream water level at the i-th time node in the integral time-delay result sequence, l' d,i is the downstream water level at the i-th time node in any sequence group in the historical sequence group set;

[0052] The most similar upstream water level sequence in history is calculated through the most similar downstream water level sequence in history. The calculation formula is as follows:

[0053]

[0054] In the formula, is the downstream water level of the i-th time node of the most similar downstream water level sequence in the historical sequence group set, is the upstream water level at the i-th time node of the most similar upstream water level sequence in the historical sequence group set;

[0055] The upstream water level of the river channel is predicted by the most similar upstream water level in history. The calculation formula is as follows:

[0056]

[0057] In the formula, l u,i is the prediction result of the upstream water level of the canal, and β is the prediction correction coefficient of the upstream water level.

[0058] Furthermore, the calculation formula of the prediction correction coefficient of the upstream water level is as follows:

[0059]

[0060] Example 2: This example takes the Xuhong River Suining Station-Pizhou Station section of the East Route of the South-to-North Water Diversion Project in Jiangsu Province as an example, and predicts the upstream water level of the channel according to the method described in the patent. In order to reflect the superiority of the proposed method for predicting the upstream water level of the channel, a fusion prediction model of a similarity prediction model and an integral time-lag model is constructed, and a general similarity prediction model is selected as a comparison method for the example. By comparing the effects of the general similarity prediction model and the proposed method, the implementation process and effect of the invention are described.

[0061] (1) Source of upstream water level prediction dataset and parameter setting

[0062] The historical channel and pond upstream and downstream water level monitoring sequence uses hourly data from May 16 to March 17, 2023, and the test set uses hourly data from April 11 to April 26, 2023. The results after the implementation of the technical solution are shown in Table 1. Figure 1 , Figure 2 , Figure 3 , Figure 4 .

[0063] (2) Water level prediction upstream of canal pond

[0064] The steps of using the proposed method to predict the water level upstream of the canal are as follows:

[0065] (2-1) Use historical data sets to construct similar prediction historical sequence groups;

[0066] (2-2) Construct an integrated time-lag model to predict the downstream water level of the canal pond in the future period;

[0067] (2-3) The downstream water level sequence of the canal pool obtained by the integrated time-lag prediction model is used as the correlation downstream water level sequence of the similarity analysis model. The similarity prediction model is constructed through the Manhattan distance to predict the upstream water level of the canal pool under the same historical scenario.

[0068] (3) Comparative analysis of the upstream water level prediction results of the canal pond using similar predictions and the proposed method

[0069] It can be seen from Table 1 that the average RMSE, average MAPE, average MSE and average MAE of the proposed method for predicting the upstream water level of the canal pool are lower than those of similar prediction methods, indicating that the average level of deviation of the prediction results of the upstream water level of the canal pool of the proposed method is smaller and the average accuracy is higher.

[0070] Depend on Figure 2 , 3 , 4, and 5, it can be seen that the RMSE, MAPE, MSE, and MAE of the water level upstream of the canal and pond predicted by the method proposed in the present invention are generally lower than the results of similar prediction methods, indicating that the overall level of deviation of the canal and pond upstream water level prediction results of the proposed method is smaller and the overall accuracy is higher.

[0071] Compared with the similarity prediction method, the proposed method takes the prediction results of the integral time-lag model as the related sequence of the similarity prediction model, fully utilizes the advantages of the predictability of the integral time-lag model and the high efficiency of the similarity prediction model, overcomes the uncertainty of the related sequence deviation of the similarity prediction model, and significantly improves the prediction accuracy of the water level upstream of the canal.

[0072] Table 1 Comparison of prediction results between the proposed method and similar prediction methods

[0073]

[0074] The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. A method for predicting upstream water level of a canal pool based on a data mechanism fusion model, characterized by: It includes the following steps: Step 1: Use historical data sets to construct similar prediction historical sequence groups; Step 2: Construct an integrated time-lag model to predict the downstream water level of the canal pond in the future period; Step 3: The downstream water level sequence of the canal pool obtained by the integrated time-lag prediction model is used as the correlation downstream water level sequence of the similarity analysis model. A similarity prediction model is constructed through Manhattan distance to predict the most similar downstream water level of the canal pool, determine the historical scenario at the same time, and predict the upstream water level of the canal pool under the same historical scenario.

2. The method for predicting upstream water level of a canal and pond based on a data mechanism fusion model according to claim 1 is characterized by: The step 1 specifically includes the following contents: First, the historical river and canal water delivery data is cleaned, including the upstream flow, downstream flow, diversion flow, and downstream water level data of the river and canal, to ensure the integrity and accuracy of the data at each time point, and obtain the upstream flow, downstream flow, diversion flow, and downstream water level sequence set with a total time length of L; Define the relevant sequence time length of the historical similar scenes of the similar prediction model as p, select the upstream water level and downstream water level long sequence historical data of the canal as the similar prediction historical data set, and divide the upstream water level and downstream water level long sequence historical data into N similar prediction historical sequence group sets according to the length of p, where each sequence group contains the upstream water level sequence and downstream water level sequence of length p; The total number of similar prediction sequence groups is calculated as follows: N = L-p+1; Where L is the total time length of the historical data set.

3. The method for predicting upstream water level of a canal or pond based on a data mechanism fusion model according to claim 1, characterized in that: The step 2 specifically includes the following contents: Taking p+1 upstream flows, downstream flows, diversion flows and initial downstream water levels in the future period as input, the next P downstream water levels are predicted to obtain a downstream water level sequence of the channel pool with a length of p. The specific prediction formula is as follows: Where s(t) is the water depth H downstream of the canal at time t. d (t) relative to the downstream water depth H at the initial time d (0), q in (t-τ) is the inflow flow rate Q of the canal pool at time (t-τ) in (t-τ) relative to the inflow flow q at the initial time in (0), q out (t) represents the outflow rate Q of the canal pool at time t out (t) relative to the outflow flow rate Q at the initial moment out (0), d(t) represents the increment of the diversion outlet flow rate D(t) at time t relative to the diversion outlet flow rate D(0) at the initial time, and the parameters of the backwater area A and the hysteresis time τ are obtained by hydrodynamic model identification.

4. The method for predicting upstream water level of a canal and pond based on a data mechanism fusion model according to claim 1, characterized in that: The step 3 specifically includes the following contents: The downstream water level sequence predicted by the integrated time-lag model is used as the correlation downstream water level sequence data of the similarity analysis model. The Manhattan distance between the correlation downstream water level sequence and each group of downstream water level sequences in the similar prediction historical sequence group is calculated. The historical downstream water level sequence with the smallest Manhattan distance is taken as the most similar historical downstream water level sequence. The Manhattan distance calculation formula is as follows: In the formula, l d,i is the downstream water level at the i-th time node in the integral time-delay result sequence, l' d,i is the downstream water level at the i-th time node in any sequence group in the historical sequence group set; The most similar upstream water level sequence in history is calculated through the most similar downstream water level sequence in history. The calculation formula is as follows: In the formula, is the downstream water level of the i-th time node of the most similar downstream water level sequence in the historical sequence group set, is the upstream water level at the i-th time node of the most similar upstream water level sequence in the historical sequence group set; The upstream water level of the river channel is predicted by the most similar upstream water level in history. The calculation formula is as follows: In the formula, l u,i is the prediction result of the upstream water level of the canal, and β is the prediction correction coefficient of the upstream water level.

5. The method for predicting upstream water level of a canal and pond based on a data mechanism fusion model according to claim 4 is characterized by: The calculation formula of the prediction correction coefficient of the upstream water level is as follows:

Images