Qualitative and Quantitative Evaluation Method for the Effect of Water Temperature Ecological Regulation Based on Deep Learning

Abstract

The present invention discloses a qualitative and quantitative evaluation method for the effect of water temperature ecological regulation based on deep learning. Based on the vertical temperature distribution data, the piecewise cubic Hermite interpolation method is used to calculate the equivalent water intake elevation, which characterizes the water intake elevation position under different regulation methods. Based on the random forest model, the mapping relationship between hydrodynamic parameters and water intake elevation is trained, and the model is optimized based on the MSE, NSE, and MAE indicators. The Shapley value method is used to analyze the key variables affecting the water intake elevation, and the contribution degree of each hydrodynamic factor to the response of ecological regulation is quantified. When other conditions remain unchanged during the ecological regulation period, the state of the stoplog gate is set to 0, and the change amount of the water intake elevation without ecological regulation is calculated. According to the water intake elevation result without ecological regulation, the predicted value of the tail water temperature under the counterfactual framework is obtained, and combined with the uncertainty analysis method, the improvement effect of ecological regulation is quantitatively evaluated.

Description

Technical Field

[0001] The present invention relates to the technical fields of data science and statistics, and particularly to a qualitative and quantitative evaluation method for the effect of water temperature ecological regulation based on deep learning. Background Art

[0002] The evaluation of the effect of water temperature ecological regulation involves a variety of complex non-linear relationships, and multiple data sources need to be comprehensively considered. The tail water temperature is only determined by the intake elevation and has nothing to do with the absolute magnitude of the upstream temperature. Since the tail water intake elevation is affected by hydrodynamic conditions such as the water level and flow rate in front of the dam and the state of the stoplog gate, it has strong non-linearity and uncertainty. Traditional models are difficult to accurately describe the relationship between the intake elevation and variables, while deep learning can capture complex non-linear relationships through multi-layer network structures to meet actual needs. In addition, the traditional method calculates the improvement effect of the stoplog gate by interpolation method, without considering the changes of other hydrodynamic conditions before and after, so there are certain errors in the result evaluation. Summary of the Invention

[0003] Therefore, the present invention provides a qualitative and quantitative evaluation method for the effect of water temperature ecological regulation based on deep learning. By applying the knowledge in the hydrological field and combining deep learning with optimization algorithms, the qualitative and quantitative evaluation of the improvement effect of water temperature ecological regulation is carried out to solve the problems proposed in the background art.

[0004] To achieve the above object, the present invention provides the following technical solution: A qualitative and quantitative evaluation method for the effect of water temperature ecological regulation based on deep learning, comprising:

[0005] Step 1: Data collection: Obtain the time series of the water level in front of the dam, the time series of the flow rate, the vertical temperature distribution data, and the state information of the stoplog gate to form a reservoir hydrodynamic database;

[0006] Step 2: Calculate the intake elevation: Based on the vertical temperature distribution data, use the piecewise cubic Hermite interpolation method to calculate the equivalent intake elevation to characterize the intake elevation position under different regulation methods;

[0007] Step 3: Deep learning modeling: Based on the random forest model, train the mapping relationship between hydrodynamic parameters and the intake elevation, and optimize the model based on the mean square error MSE, Nash efficiency coefficient NSE, and mean absolute error MAE indicators;

[0008] Step 4: Qualitative evaluation: Use the Shapley value method to analyze the key variables affecting the intake elevation and quantify the contribution degree of each hydrodynamic factor to the ecological regulation response;

[0009] Step 5: Counterfactual estimation: When other conditions remain unchanged during the ecological regulation period, set the state of the stoplog gate to 0 and calculate the change amount of the intake elevation without ecological regulation.

[0010] Step 6: Quantitative evaluation: Based on the water intake elevation results under the non-ecological operation, the predicted value of the tail water temperature under the counterfactual framework is further obtained, and the improvement effect of the ecological operation is quantitatively evaluated by combining the uncertainty analysis methods of mean, variance, and maximum probability point.

[0011] Preferably, the data collected in Step 1 are specifically: the time series of the water level in front of the dam , the time series of the flow rate , the vertical temperature distribution data , the tail water temperature , , is the total number of time sampling points; where: represents the water level height corresponding to a certain moment in front of the dam; represents the inflow corresponding to a certain moment in front of the dam; represents the change of the water temperature distribution at different depths in front of the dam over time ; represents the water temperature corresponding to a certain moment of the tail water; the ecological operation information of the water intake in front of the dam is collected, that is, the status information of the stop logs , where: represents the stop logs being closed; represents the stop logs being opened;

[0012] Using Z-score to detect outliers: Using the standardization formula , where represents the time series of the water level in front of the dam at time , the time series of the flow rate , the vertical temperature distribution data , the tail water temperature , where and represent the mean and standard deviation of to respectively, represents the value of each item at time , represents the value of each item at time ; the outlier determination method is ;

[0013] Processing missing values or outliers: For the monitoring data with missing or abnormal situations, interpolation of missing values and outliers based on time series autoregressive analysis is performed; the following formula is used:

[0014] ;

[0015] ​​Among them represents the time history of the water level in front of the dam , the time history of the flow rate , the vertical temperature distribution data , the tail water temperature at the time for missing values or outliers; represents the time history of the water level in front of the dam , the time history of the flow rate , the vertical temperature distribution data , the tail water temperature at the time for known monitored values, , are autoregressive and moving average parameters; is the intercept term, is the autoregressive order, is the moving average parameter, is the error term.

[0016] Preferably, step 2 is specifically: in the vertical temperature distribution data , assume that there are known known depth points corresponding to the water temperature , among which , are the known depth and the temperature corresponding to the depth respectively; the basic formula for piecewise cubic Hermite interpolation is as follows:

[0017] ;

[0018] Among them,

[0019] ,

[0020] ,

[0021] ,

[0022] ,

[0023] ,

[0024] and are the derivatives at the interpolation points and , calculated by finite differences;

[0025] Use cubic Hermite interpolation to establish the continuous mapping relationship between the water temperature in front of the dam and the depth : Set the tail water temperature , the Newton iteration method is used to solve the elevation of the intake before the dam corresponding to the tail water , that is

[0026] ;

[0027] Among them is the derivative of the cubic Hermite interpolation function, is t the elevation obtained at the th iteration at time ; when .

[0028] Preferably, step 3 is specifically: select the time history of the water level before the dam , the time history of the flow , and the ecological regulation state as the input variables of deep learning, and the output variable is the elevation of the tail water intake ; Random Forest Regression (RF) is based on the decision tree integration method and is used to establish the mapping relationship between hydrodynamic conditions and the intake elevation. Its basic model:

[0029] ;

[0030] Among them, represents the parameter set of the random forest model, including the number of trees , the maximum depth of the tree and the training set division ratio ; The parameters are determined by the optimization method, and the optimization selection evaluation indicators are: mean square error ; Nash efficiency coefficient , mean absolute error ; among them is the actual value, is the predicted value, is the average value of the actual values, n is the number of samples, NSE The closer the value of is to 1, the better the model fitting effect; the predicted intake elevation output of the random forest:

[0031] ;

[0032] Among them is the prediction result of the th decision tree.

[0033] Preferably, the specific steps of step 4 are:

[0034] During the training of the deep learning random forest and obtaining the prediction result Based on this, for each input feature of the water level in front of the dam, the flow time history, and the ecological regulation state, perform multiple random permutations to generate multiple different feature subsets;

[0035] For each feature subset, use the trained model for prediction and calculate its prediction result for the tail water intake elevation where , is the total number of Monte Carlo simulations;

[0036] For each input variable , which represents one of the variables among the water level in front of the dam, the flow rate, and the ecological regulation state, calculate its Shapley value, indicating the contribution of this variable to the prediction result of the tail water intake elevation:

[0037] ;

[0038] where: is the feature subset without for the th random sampling; is the model prediction result corresponding to the feature subset at time t, indicating the th random sampling of the feature subset containing .

[0039] Preferably, during the ecological regulation period , set a counterfactual scenario, that is, assume no flap gate state during this time period , while keeping the water level time history in front of the dam , the flow rate time history and other conditions unchanged. Use the trained deep learning random forest model. Under the new input conditions , with other conditions unchanged, substitute into the formula to recalculate the intake elevation in the counterfactual situation. The calculation result is denoted as , that is:

[0040] .

[0041] Preferably, in the case where the counterfactual intake elevation result is known, substitute it into the above Hermite interpolation method result:

[0042] ;

[0043] Calculate the predicted value of the counterfactual tail water temperature and calculate its difference:

[0044] ;

[0045] Use the mean value to measure the effect of ecological regulation:

[0046] ;

[0047] If , it indicates that the water temperature without ecological regulation is higher than the current ecological regulation;

[0048] Use the variance to measure the volatility of the water temperature difference:

[0049] ;

[0050] A smaller variance means that the volatility of the water temperature difference over time is smaller, reflecting the impact of ecological regulation on water temperature stability;

[0051] Use the most probable point MPP to represent the most common value of the water temperature difference:

[0052] ;

[0053] where is the probability density function PDF of the water temperature difference.

[0054] The present invention has the following advantages:

[0055] The present invention constructs a random forest model to establish the mapping relationship between hydrodynamic conditions and water intake elevation. Through a single control variable, it can calculate the change in water intake elevation brought about by ecological regulation under the condition that other hydrodynamic conditions remain unchanged, and then more accurately analyze the impact of the stoplog gate on water temperature; at the same time, through the counterfactual prediction method combined with the mean value, variance, and most probable point uncertainty quantification analysis method, a quantitative evaluation of the improvement effect of water temperature ecological regulation is carried out. Compared with traditional methods, it can adapt to high-dimensional and non-linear data, and the results are more accurate and objective. Brief Description of the Drawings

[0056] Figure 1 is the flow chart provided by the present invention;

[0057] Figure 2 is the fitting result graph of Reservoir A using deep learning in 2024;

[0058] Figure 3 is the SHAP value result graph of Reservoir A in 2024;

[0059] Figure 4 is the comparison graph of counterfactual prediction data of the tail water temperature of Reservoir A in April 2024;

[0060] Figure 5It is a comparison chart of counterfactual prediction data of the water temperature distribution at the tail water of Reservoir A in April 2024. Detailed implementation manners

[0061] The following specific embodiments illustrate the implementation manners of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0062] Embodiment: The biodiversity of aquatic organisms in the lower reaches of a certain river is relatively high. The change of water temperature in the downstream river channel will interfere with the fish reproduction and have an important impact on the river water ecological environment. Four cascade reservoirs planned in the lower reaches of a certain river, namely Reservoir A, Reservoir B, Reservoir C, and Reservoir D, started to store water and operate in January 2020, April 2021, May 2013, and October 2012 respectively. The operation of the reservoirs will change the temporal and spatial distribution of the natural river runoff in the downstream, have an important impact on the water temperature distribution and seasonal variation along the river channel, resulting in obvious flattening and delay effects of the river water temperature. Moreover, the cumulative effect will also be produced due to the change of water temperature caused by cascade hydropower development.

[0063] Theoretically, by opening and closing the stoplogs to take the upper and middle layer water, the adverse effects of the warming delay on fish spawning drifting eggs and fish spawning adhesive sinking eggs can be alleviated. The ecological regulation of Reservoir A, Reservoir B, and Reservoir C is to achieve the upper and middle layer water intake of the corresponding reservoirs by lowering the stoplogs, increase the water temperature at the reservoir outlet, and enable the river section below the dam to reach the water temperature suitable for the spawning and reproduction of fish spawning drifting eggs and fish spawning adhesive sinking eggs earlier. The present invention conducts a modeling analysis on the response relationship between hydrodynamic factors and water intake elevation during the ecological regulation period, and uses counterfactual prediction to analyze the improvement effect of the stoplog stratified water intake on water temperature during the ecological regulation period.

[0064] Therefore, the present invention provides a qualitative and quantitative evaluation method for the water temperature ecological regulation effect based on deep learning, as Figure 1 shown, including the following steps:

[0065] Step 1, obtain the time series of the water level in front of the dam, the time series of the flow rate, the vertical temperature distribution data, and the stoplog status information to form a reservoir hydrodynamic database, including:

[0066] Collect the time series of the water level in front of the dam and the time series of the flow rate and the vertical temperature distribution data and the tail water temperature , , where is the total number of time sampling points; among them: represents the water level in front of the dam at a certain moment The corresponding water level height; Indicates the inflow corresponding to a certain moment in front of the dam ; Indicates the change over time of the water temperature distribution at different depths in front of the dam ; Change; Indicates the water temperature corresponding to a certain moment in the tail water Ecological regulation information of the water intake in front of the dam is collected, that is, the status information of the stoplogs , where: Indicates that the stoplogs are closed; Indicates that the stoplogs are open, is the total number of time sampling points.

[0067] Use Z-score to detect outliers. Use the standardization formula , where represents the time series of the water level in front of the dam at time , the time series of the flow rate , the vertical temperature distribution data , the tail water temperature , where and respectively represent to the mean and standard deviation, represents the values of each item at time represents the values of each item at time. The outlier determination method is .

[0068] Process missing values or outliers. For the monitoring data with missing or abnormal conditions, perform interpolation of missing values and outliers based on time series autoregressive analysis. Use the following formula:

[0069] ;

[0070] where represents the missing value or outlier in the time series of the water level in front of the dam , the time series of the flow rate , the vertical temperature distribution data , the tail water temperature at time ; represents the known monitoring values in the time series of the water level in front of the dam , the time series of the flow rate , the vertical temperature distribution data , the tail water temperature at time , , are the autoregressive and moving average parameters; is the intercept term, is the autoregressive order, are the moving average parameters, is the error term.

[0071] Step 2: Based on the vertical temperature distribution data, the piecewise cubic Hermite interpolation method is used to calculate the equivalent water intake elevation, which characterizes the water intake elevation position under different scheduling methods, including:

[0072] Among the water temperature data in front of the dam let there be known known depth points corresponding to water temperatures where , are the known depth and the temperature at the corresponding depth respectively. The basic formula for piecewise cubic Hermite interpolation is as follows:

[0073] ;

[0074] where

[0075] ,

[0076] ,

[0077] ,

[0078] ,

[0079] ,

[0080] and are the derivatives at the interpolation points and and are calculated by finite differences.

[0081] Furthermore, cubic Hermite interpolation is used to establish a continuous mapping relationship between the water temperature in front of the dam and the depth : Set the tail water temperature , and use the Newton iteration method to solve the water intake elevation in front of the dam corresponding to the tail water, that is

[0082] ;

[0083] where is the derivative of the cubic Hermite interpolation function, is t the elevation obtained at the th iteration at time When 。

[0084] Step 3: Use the deep learning random forest model method to establish the mapping relationship between hydrodynamic conditions and intake elevation, and optimize the model based on indicators such as mean square error (MSE), Nash efficiency coefficient (NSE), and mean absolute error (MAE), including:

[0085] Select the time series of the water level in front of the dam as the deep learning input variable 、the time series of flow rate 、and the ecological operation status , and the output variable is the tailwater intake elevation ; Random Forest Regression (RF) is based on the decision tree integration method and is used to establish the mapping relationship between hydrodynamic conditions and intake elevation. Its basic model:

[0086] ;

[0087] Among them, represents the parameter set of the random forest model, including the number of trees 、the maximum depth of the tree and the training set division ratio 。 The parameters are determined by the optimization method, and the optimization selection evaluation indicators are: mean square error MSE, Nash efficiency coefficient NSE, and mean absolute error MAE. Mean square error ; Nash efficiency coefficient , mean absolute error ; where is the actual value, is the predicted value, is the average value of the actual values, n is the number of samples, NSE The closer the value of is to 1, the better the model fitting effect. The predicted intake elevation output of the random forest:

[0088] ;

[0089] Among them is the prediction result of the th decision tree.

[0090] Step 4: Use the Shapley value method to analyze the contributions of the time series of the water level in front of the dam, the time series of flow rate, and the ecological operation status to the tailwater intake elevation, and reveal the key driving factors, including:

[0091] When training the deep learning random forest and obtaining the prediction results ( , On the basis of the total number of time sampling points, for each input feature (the water level in front of the dam, the flow time history, and the ecological regulation status), multiple random permutations are performed to generate multiple different feature subsets. For each feature subset, the trained model is used for prediction, and its prediction result for the tail water intake elevation is calculated, where , is the total number of Monte Carlo simulations.

[0092] For each input variable (representing one of the variables of the water level in front of the dam, the flow rate, and the ecological regulation status), its Shapley value is calculated, indicating the contribution of this variable to the prediction result of the tail water intake elevation:

[0093] ;

[0094] where: is the feature subset without in the th random sampling; is the model prediction result corresponding to the feature subset at time t, represents the feature subset containing in the th random sampling.

[0095] Step 5, perform counterfactual estimation, including:

[0096] During the ecological regulation period ( ), set the counterfactual scenario. That is, assume no flap gate state in this time period, while keeping the water level time history in front of the dam , the flow time history unchanged, etc. Use the trained deep learning random forest model. Under the new input conditions , with other conditions unchanged, substitute into the formula , and recalculate the intake elevation in the counterfactual situation. The calculation result is recorded as , that is:

[0097] .

[0098] Step 6, quantitatively evaluate the results, including:

[0099] According to the intake elevation result without ecological regulation, further obtain the predicted value of the tail water temperature in the counterfactual framework, and combine the mean value, variance, and maximum probability point uncertainty analysis methods to quantitatively evaluate the improvement effect of ecological regulation. In the counterfactual intake elevation result Substitute the known conditions into the result of the Hermite interpolation method described in step 3:

[0100] ;

[0101] Calculate the predicted value of the counterfactual tailwater temperature , and calculate its difference:

[0102] ;

[0103] Use the mean value to measure the effect of ecological regulation:

[0104] ;

[0105] If , it indicates that the water temperature without ecological regulation is higher than the current ecological regulation.

[0106] Use the variance to measure the volatility of the water temperature difference:

[0107]

[0108] A smaller variance means that the volatility of the water temperature difference over time is smaller, reflecting the impact of ecological regulation on water temperature stability.

[0109] Use the maximum probability point (MPP) to represent the most common value of the water temperature difference:

[0110] ;

[0111] where is the probability density function (PDF) of the water temperature difference.

[0112] Now use this invention to analyze the ecological regulation data of Reservoir A.

[0113] In step s1, obtain the water level before the dam of Reservoir A, the inflow rate, the outflow rate, and the water temperature data every hour, and use to detect outliers and use the ARIMA model for interpolation;

[0114] In step s2, use the obtained hydrodynamic data to establish and train a random forest regression model, evaluate the model, and obtain the fitting result; use the model to obtain the response relationship between each hydrodynamic parameter and the water intake elevation, and obtain that the mean squared error MSE of the evaluation result is 0.009 and the R² value is 0.998, indicating that the model fitting effect is good.

[0115] In step s3, analyze the key variables affecting the water intake elevation by calculating the SHAP value;

[0116] From Figure 2Comparison results between the predicted values and the true values of the random forest regression model. The horizontal axis represents the true values, and the vertical axis represents the predicted values of the model. The blue scatter points represent the prediction results of the model, and the red dashed line is the ideal prediction line. From Figure 2 it can be seen that the data points are distributed near the ideal prediction line, indicating that the prediction results of the random forest regression model are in good agreement with the true values.

[0117] From Figure 3 it is known that the SHAP value distribution of the variable "whether to start scheduling" is relatively dense, and the median is greater than zero, that is, whether to carry out ecological scheduling has a significant impact on the water intake elevation. During the ecological scheduling period, the water intake elevation has a significant increase.

[0118] Step S4, through deep learning counterfactual prediction, set the stoplog gate state to 0, and use the output water intake elevation to inversely obtain the corresponding predicted value of the tailrace water temperature to quantitatively analyze the improvement effect of ecological scheduling;

[0119] From Figure 4 , the range of the tailrace water temperature is between 14.5 degrees and 18 degrees. The ecological scheduling observation value (blue line) is slightly higher than the counterfactual prediction value (red line) as a whole, with an increase of between 0.5 degrees and 1.0 degrees. That is, during the ecological scheduling period, the tailrace water temperature has increased significantly, which can effectively alleviate the adverse effects of warming delay on fish species that produce drifting eggs and fish species that produce sticky and sinking eggs.

[0120] Figure 5 Shows the comparison of the water temperature distribution histogram and the kernel density estimation curve between the ecological scheduling observation value and the counterfactual prediction value. The water temperature distribution of the ecological scheduling observation value is slightly shifted to the right. The peak of the water temperature distribution of the counterfactual prediction value is relatively shifted to the left. It shows that under the condition of ecological scheduling, the peak of the probability density of the tailrace water temperature has increased significantly.

[0121] Step S5, combine the mean, variance, and maximum probability point uncertainty analysis methods to evaluate the effect during the ecological scheduling period of Reservoir A;

[0122] Table 1

[0123]

[0124] From Table 1, it can be found from the data in the table that the present invention can evaluate the improvement effect from multiple dimensions. In terms of the mean, the water temperature improvement effect is significant; in terms of the variance, the water temperature improvement effect is more stable during the ecological scheduling period. It can be found that ecological scheduling can effectively increase the water temperature of the downstream river channel and improve the aquatic ecological environment of the river channel. Compared with the present invention, the traditional method can only give a general statement about the water temperature improvement effect before and after the presence or absence of the stoplog gate. Therefore, this method can not only improve the accuracy of result analysis but also make up for the shortcoming of its single analysis angle.

[0125] The basic principle of the method is that the temperature of the tail water is only determined by the intake elevation and has nothing to do with the absolute magnitude of the upstream temperature. Since the intake elevation of the tail water is affected by hydrodynamic conditions such as the water level and flow rate in front of the dam and the state of the stoplog gate, it has strong nonlinearity and uncertainty. It is difficult for traditional models to accurately describe the relationship between the intake elevation and variables, while deep learning can capture complex nonlinear relationships through multi-layer network structures to meet actual needs. In addition, the traditional method calculates the improvement effect of the stoplog gate through interpolation and fails to consider changes in other hydrodynamic conditions before and after, so there are certain errors in the result evaluation. By establishing a model and controlling a single variable, the result is made more accurate and the evaluation method is further optimized. The present invention constructs the response relationship between each hydrodynamic parameter and the intake elevation through deep learning, then quantifies the contribution of each feature to the prediction result through the Shapley method, and comprehensively evaluates the influence effect of the stoplog gate on the intake elevation during ecological regulation according to the counterfactual prediction combined with the quantitative evaluation methods of uncertainty quantification of mean, variance, and maximum probability point.

[0126] Although the present invention has been described in detail above with general descriptions and specific embodiments, on the basis of the present invention, some modifications or improvements can be made, which are obvious to those skilled in the art. Therefore, these modifications or improvements made without departing from the spirit of the present invention all fall within the scope of protection required by the present invention.

Claims

1. A qualitative and quantitative evaluation method for the effect of water temperature ecological regulation based on deep learning, characterized in that: Including: Step 1: Obtain the time series of the water level in front of the dam, the time series of the flow rate, the vertical temperature distribution data, and the status information of the stop logs to form a reservoir hydrodynamic database; Step 2: Based on the vertical temperature distribution data, use the piecewise cubic Hermite interpolation method to calculate the equivalent water intake elevation, which characterizes the water intake elevation position under different operation modes; Step 3: Based on the random forest model, train the mapping relationship between hydrodynamic parameters and the water intake elevation, and optimize the model based on the mean square error MSE, Nash efficiency coefficient NSE, and mean absolute error MAE indicators; Step 3 specifically is: select the input variables of deep learning, namely the water level time series H(t) in front of the dam, the flow rate time series Q(t), and the ecological operation state S damper , and the output variable is the tailwater intake elevation Z o (t); Random forest regression is based on the decision tree ensemble method and is used to establish the mapping relationship between hydrodynamic conditions and the intake elevation. Its basic model: Among them, Θ represents the parameter set of the random forest model, including the number of trees N t , the maximum depth d of the tree, and the training set division ratio α; the Θ parameter is determined by an optimization method, and the optimization selects evaluation indicators: mean square error ; Nash efficiency coefficient , mean absolute error ; where Z i is the actual value, is the predicted value, is the average value of the actual values, n is the number of samples, and the closer the NSE value is to 1, the better the model fitting effect; the predicted water intake elevation output of the random forest: wherein is the prediction result of the i-th decision tree; Step 4: Use the Shapley value method to analyze the key variables affecting the water intake elevation and quantify the contribution degree of each hydrodynamic factor to the response of ecological operation; Step 5: When other conditions remain unchanged during the ecological operation period, set the status of the stop logs to 0 and calculate the change in the water intake elevation without ecological operation; Step 6: According to the results of the water intake elevation without ecological operation, further obtain the predicted value of the tail water temperature under the counterfactual framework, and combine the uncertainty analysis method to quantitatively evaluate the improvement effect of ecological operation.

2. The qualitative and quantitative evaluation method for the water temperature ecological regulation effect based on deep learning according to claim 1, characterized in that: The data collected in Step 1 are specifically: the time series of the water level in front of the dam H(t), the time series of the flow rate Q(t), the vertical temperature distribution data T(z,t), and the tail water temperature T o (t), where t = 1, 2, … N, and N is the total number of time sampling points; among them: H(t) represents the water level height corresponding to a certain moment t in front of the dam; Q(t) represents the inflow corresponding to a certain moment t in front of the dam; T(z,t) represents the change of the water temperature distribution at different depths z in front of the dam with time t; T o (t) represents the water temperature corresponding to a certain moment t in the tail water; collect the ecological regulation information of the water intake in front of the dam, that is, the status information S of the stoplog damper , where: S damper = 1 indicates that the stoplog is closed; S damper = 0 indicates that the stoplog is open; Using Z-score to detect outliers: Using the normalization formula , where X t represents the time series of the water level in front of the dam H(t), the time series of the flow rate Q(t), the vertical temperature distribution data T(z,t), and the tail water temperature T o (t), where μ and σ represent the mean and standard deviation of X t-10 to X t-1 , respectively, X t-10 represents the values at time t-10, and X t-1 represents the values at time t-1; Process missing values or outliers: For missing or abnormal monitoring data, perform interpolation of missing values and outliers based on time series autoregressive analysis; use the following formula: where X t represents the missing values or outliers at time t in the time series of the water level in front of the dam H(t), the time series of the flow rate Q(t), the vertical temperature distribution data T(z,t), and the tailwater temperature T o (t); X t-i represents the known monitored values at time t-i in the time series of the water level in front of the dam H(t), the time series of the flow rate Q(t), the vertical temperature distribution data T(z,t), and the tailwater temperature T o (t), φ i , θ j are the autoregressive and moving average parameters; c is the intercept term, p is the autoregressive order, q is the moving average parameter, and ε t is the error term.

3. The qualitative and quantitative evaluation method for the water temperature ecological regulation effect based on deep learning according to claim 1, characterized in that: Step 2 specifically is: In the vertical temperature distribution data T(z,t), assume that there are n known depth points corresponding to water temperatures (z1,T1(t)), (z2,T2(t)),...,(z n ,T n (t)), where z i ,T i are the known depths and the temperatures corresponding to the respective depths; the basic formula for piecewise cubic Hermite interpolation is as follows: T H (z) = h0(z)T i (t) + h1(z)T i+1 (t) + h2(z)m i + h3(z)m i+1 , z i ≤ z ≤ z i+1 ; where, h0(z) = 1 - 3s 2 + 2s 3 , h1(z) = 3s 2 -2s 3 , h2(z) = (s - 2s 2 + s 3 )·(z i+1 - z i ), h3(z) = (-s 2 +s 3 )·(z i+1 -z i ), m i and m i+1 are the derivatives at the interpolation points z i and z i+1 respectively, calculated by finite differences; A continuous mapping relationship between the water temperature T(z, t) in front of the dam and the depth z is established by using cubic Hermite interpolation: The tail water temperature T o (t) is set, and the Newton iteration method is used to solve the elevation Z o (t) of the water intake in front of the dam corresponding to the tail water, that is where T′ H (z (k) (t),t) is the derivative of the cubic Hermite interpolation function, and z (k+1) (t) is the corresponding elevation obtained at the (k + 1)-th iteration at time t.

4. The qualitative and quantitative evaluation method for the effect of water temperature ecological regulation based on deep learning according to claim 1, characterized in that: The specific steps of Step 4 are as follows: On the basis of training a deep learning random forest and obtaining the prediction result HF(H(t), Q(t), S damper (t); Θ), for each input feature of the water level in front of the dam, the flow time history, and the ecological regulation state, perform multiple random permutations to generate multiple different feature subsets; For each feature subset, use the trained model to make predictions and calculate its predicted results for the elevation of tail water intake where m = 1, 2, … M, and M is the total number of Monte Carlo simulations; For each input variable X i , X i represents one of the variables of the water level in front of the dam, flow rate, and ecological regulation status, and calculates its Shapley value, φ i represents the contribution of this variable to the prediction result of the tail water intake elevation: Where: S m is the feature subset without X i for the m-th random sampling; HF(S m , t) is the model prediction result corresponding to the feature subset S m at time t, and S m ∪{X i} represents the feature subset with X i for the m-th random sampling.

5. The qualitative and quantitative evaluation method for the water temperature ecological regulation effect based on deep learning according to claim 4, characterized in that: During the ecological operation period S damper = 1, a counterfactual scenario is set, that is, it is assumed that the state of the stoplog gate S damper = 0 during this time period, while keeping the time series of the water level H(t) and the flow rate Q(t) in front of the dam unchanged. Using the trained deep learning random forest model, under the new input condition S damper = 0 and other conditions remaining unchanged, substitute it into the formula HF(H(t), Q(t), S damper ; Θ), and recalculate the water intake elevation HF(H(t), Q(t), 0; Θ) in the counterfactual situation. The calculation result is denoted as That is:

6. The qualitative and quantitative evaluation method for the water temperature ecological regulation effect based on deep learning according to claim 5, characterized in that: In the counterfactual water intake elevation result For the known situation, substitute it into the result of the Hermite interpolation method described in claim 3: Calculate the predicted value T H,C (t) of the counterfactual tailwater temperature, and calculate its difference: ΔT o (t) = T H,C (t) - T o (t), t = 1, 2, … N; Use the mean value to measure the effect of ecological operation: If It shows that the water temperature without ecological operation is higher than that under the current ecological operation; Use the variance to measure the volatility of the water temperature difference: Use the most probable point MPP to represent the most common value of the water temperature difference: MPP(ΔT o ) = argmax P(ΔT o ); where P(ΔT o ) is the probability density function (PDF) of the water temperature difference.

Images