Reservoir water temperature cumulative impact assessment method and device based on artificial intelligence

Abstract

A reservoir water temperature cumulative impact evaluation method and device based on artificial intelligence are disclosed. The method comprises the following steps: decomposing the air temperature time series of the river section where the target reservoir system is located, and removing the long-term warming trend item to obtain a de-trended air temperature sequence; using a trained artificial intelligence time series model to respectively reconstruct a dam-free scenario water temperature sequence and a long-term warming-free scenario water temperature sequence; forming a three-scenario water temperature sequence system by combining the downstream actual observed water temperature sequence with the two reconstructed sequences, and performing multi-factor attribution decomposition based on the internal statistical difference relationship to decompose the total water temperature change into three independent contribution items, namely, artificial regulation, long-term climate warming, and annual climate variability; and finally, based on the results of the multi-factor attribution decomposition, performing cumulative effect operation to output the cross-season heat accumulation diagnosis result. The scheme realizes the multi-factor fine independent quantitative attribution and the system diagnosis of the water temperature heat accumulation effect.

Description

Technical Field

[0001] This disclosure relates to the field of water temperature process analysis and impact assessment technology, and in particular to a method and equipment for assessing the cumulative impact of reservoir water temperature based on artificial intelligence. Background Technology

[0002] River water temperature is a crucial factor controlling the physical environment and aquatic ecosystem functions of rivers, and also an important indicator for evaluating the ecological impact of reservoir operation. With the continuous construction of large reservoirs and cascade reservoir systems, the natural river hydrothermal processes have been significantly altered, with downstream river sections generally exhibiting seasonal warming or cooling and other thermal redistribution phenomena. Particularly in complex reservoir systems, the combined effects of multi-reservoir joint operation and thermal stratification evolution result in downstream river water temperature processes exhibiting distinct nonlinear, multi-timescale, and cross-seasonal coupling characteristics.

[0003] Existing river water temperature research mainly relies on mechanistic solution models or data-driven models that establish mapping relationships based on historical monitoring data. Some studies have also attempted to analyze the comprehensive impact of reservoir construction on water temperature using the approaches of "dam-based observation" and "damless reconstruction." However, for actual river sections with complex scheduling phenomena, existing technologies have significant limitations in attribution analysis and cumulative effect evaluation.

[0004] Existing evaluation methods typically rely on direct comparisons before and after dam construction or simplified empirical comparisons to reflect water temperature changes. This approach struggles to accurately construct a reliable natural, dam-free benchmark scenario when the target river section is under long-term regulation. Current water temperature change attribution studies often only calculate the overall change or roughly divide historical change stages, failing to further refine the complex mixed water temperature signal into the specific contribution shares of each independent driving factor. Furthermore, existing evaluation systems are mostly limited to comparisons of mean values ​​in specific seasons or descriptions of phenomena at a single time scale, lacking comprehensive linkage diagnosis and failing to reflect the cross-seasonal heat transfer and cumulative effects under complex environments.

[0005] In summary, how to reliably establish a natural water temperature reference standard in the absence of concurrent observations, and how to quantitatively decompose the water temperature contribution of different driving factors to achieve a scientific evaluation of the impact of cross-seasonal heat accumulation, are the technical challenges that urgently need to be solved in this field. Summary of the Invention

[0006] This disclosure proposes an artificial intelligence-based scheme for assessing the cumulative impact of reservoir water temperature, which can solve the technical problems of existing technologies in establishing natural reference benchmarks when there is a lack of concurrent observations, and in quantitatively decomposing the water temperature contribution of multiple driving factors and evaluating the cross-seasonal heat accumulation impact.

[0007] According to one aspect of this disclosure, an artificial intelligence-based method for assessing the cumulative impact of reservoir water temperature is proposed, including: The temperature time series of the river section where the target reservoir system is located is decomposed and the long-term warming trend term is removed to obtain the detrended temperature series; Using a pre-trained artificial intelligence time series model, the original temperature series and upstream hydrological features are used as model-driven inputs to reconstruct the water temperature series under the dam-free scenario, and the detrended temperature series and upstream hydrological features are used as model-driven inputs to reconstruct the water temperature series under the long-term warming scenario. The actual observed water temperature sequence of the downstream control station, the water temperature sequence under the damless scenario, and the water temperature sequence under the long-term warming scenario are used to form a three-scenario water temperature sequence system. Based on the statistical difference relationship between the sequences in the three-scenario water temperature sequence system, a multi-factor attribution decomposition is performed to decompose the total water temperature change into anthropogenic regulation contribution, long-term climate warming contribution, and annual climate variability contribution. Based on the results of the multi-factor attribution decomposition, the cumulative effect calculation of the water temperature process during the target evaluation period is performed to obtain the cross-seasonal heat accumulation diagnosis results.

[0008] According to one embodiment of this disclosure, an electronic device is provided, the device including a memory and a processor, the memory being used to store computer instructions executable on the processor, the processor being used to implement the method as described in any of the preceding claims when executing the computer instructions.

[0009] The proposed AI-based reservoir water temperature cumulative impact assessment scheme decomposes the air temperature time series and removes the long-term warming trend term. It then uses a pre-trained AI time series model, inputting the original air temperature series and the detrended air temperature series for model reconstruction. This achieves reliable extraction of water temperature series under both dam-free and non-long-term warming scenarios, overcoming the limitation of existing technologies in establishing reliable reference standards when contemporaneous observations are lacking. By constructing a three-scenario water temperature series system with the actual observed water temperature series and the dam-free and non-long-term warming scenarios, and performing multi-factor attribution decomposition based on the statistical difference relationships between the series, the scheme accurately decomposes the total water temperature change into contributions from human regulation, long-term climate warming, and intra-annual climate variability, significantly improving the quantitative explanation of the causes of complex water temperature changes. Based on the results of the multi-factor attribution decomposition, the scheme further calculates the cumulative effect of the water temperature process within the target assessment period, achieving accurate diagnosis of the cross-seasonal heat accumulation effect caused by complex scheduling, effectively compensating for the shortcomings of existing single-time-scale assessment methods.

[0010] Other features and advantages of this disclosure are described below. Attached Figure Description

[0011] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this specification and, together with the description, serve to explain the principles of this specification.

[0012] Figure 1 This is a flowchart of an artificial intelligence-based method for evaluating the cumulative impact of reservoir water temperature, provided in one embodiment of this disclosure; Figure 2 This disclosure provides an exemplary embodiment of a comparison of river water temperature before and after reservoir construction; Figure 3 This is a schematic diagram of the technical route of the artificial intelligence-based reservoir water temperature cumulative impact assessment method provided in an exemplary embodiment of this disclosure; Figure 4 This is a schematic diagram of the hardware structure of an electronic device provided in at least one embodiment of this disclosure. Detailed Implementation

[0013] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0014] This disclosure can be applied to computer systems / servers that can operate with a wide range of other general-purpose or special-purpose computing system environments or configurations. Examples of well-known computing systems, environments, and / or configurations suitable for use with computer systems / servers include, but are not limited to: personal computer systems, server computer systems, thin clients, thick clients, handheld or laptop devices, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputer systems, mainframe computer systems, and distributed cloud computing environments that include any of the above systems, etc.

[0015] Computer systems / servers can be described in the general context of computer system executable instructions (such as program modules) executed by the computer system. Typically, program modules can include routines, programs, object programs, components, logic, data structures, etc., which perform specific tasks or implement specific abstract data types. Computer systems / servers can be implemented in distributed cloud computing environments, where tasks are performed by remote processing devices linked through a communication network. In distributed cloud computing environments, program modules can reside on local or remote computing system storage media, including storage devices.

[0016] This disclosure provides an artificial intelligence-based method for assessing the cumulative impact of reservoir water temperature. By stripping long-term warming trends from air temperature time series data and using an AI time series model to reconstruct downstream natural water temperature processes under different climate-driven conditions, a water temperature sequence system is constructed, encompassing three scenarios: actual observation, no dam, and no long-term warming. Based on this, and using the statistical difference relationship between the three scenario sequences, the total water temperature change is decomposed into three independent contributions: human intervention, long-term climate warming, and intra-annual climate variability. Furthermore, based on the attribution results, the cumulative effect calculation of the water temperature process within the target assessment period is performed to obtain cross-seasonal heat accumulation diagnostic results, thereby quantitatively evaluating the cumulative heat impact of reservoir operation.

[0017] Please refer to Figure 1 , Figure 1 This is a flowchart of an artificial intelligence-based method for evaluating the cumulative impact of reservoir water temperature, provided in one embodiment of this disclosure. This method can be applied to river sections affected by reservoir system regulation. The target reservoir system can be a single reservoir or a cascade reservoir system consisting of two or more reservoirs. Based on upstream and downstream hydrological, water temperature, and meteorological observation data, this method performs attribution decomposition and cumulative effect evaluation on downstream water temperature changes caused by reservoir operation. As shown in the figure, the method includes the following steps S1 to S4.

[0018] Step S1: Decompose the temperature time series of the river section where the target reservoir system is located and remove the long-term warming trend term to obtain the detrended temperature series.

[0019] Before proceeding with this step, the evaluation target should be identified and a basic dataset constructed. For example, a research river section affected by the regulation of the target reservoir system can be selected to establish a data system for evaluation. This data system may include: upstream control station data located upstream of the target reservoir system and as unaffected as possible by its direct regulation; downstream control station data located downstream of the target reservoir system and capable of reflecting the comprehensive thermal response of regulation; and meteorological station data that are spatially proximate and climatically representative of the downstream control station. The basic dataset may include daily-scale flow data from the upstream control station, daily-scale water temperature data from both the upstream and downstream control stations, and daily-scale air temperature data from the meteorological station. Depending on the specific application scenario, the basic dataset may also include reservoir inflow and outflow data, water level or water storage process data, operational phase information, and other auxiliary data related to natural water temperature reconstruction and the evaluation of regulation effects. In one example, the upstream control station could be Huatan Station, the downstream control station could be Zhutuo Station, the meteorological station could be Jiangjin Station, the analysis period could be set from 2000 to 2020, and the research object could be the Xiluodu-Xiangjiaba cascade in the downstream cascade reservoir system of the Jinsha River.

[0020] The purpose of stripping the long-term warming trend from the temperature time series is to separate the multi-decadal warming component from the original temperature signal, so that the detrended temperature series used in subsequent counterfactual reconstruction retains only intra-annual and inter-annual climate variability, without carrying long-term warming signals.

[0021] In some implementations, decomposing the temperature time series and removing the long-term warming trend term includes: applying a seasonal-trend decomposition method based on local weighted regression to the temperature time series to decompose the temperature time series into a long-term trend term, a seasonal term, and a residual term; removing the long-term trend term from the temperature time series and adding back the mean of the long-term trend term during the analysis period to obtain the detrended temperature series.

[0022] Seasonal-Trend decomposition using Loess (STL) is a method that decomposes a time series into three components: a trend term, a seasonal term, and a residual term. For a diurnal temperature observation series AT(t), the STL decomposition can be expressed as: , in, The STL decomposition represents long-term warming signals over multiple decadal scales. AT_season(t) represents intra-annual cyclical variations, and AT_resid(t) represents short-term fluctuations after removing the trend and seasonal terms. Robust fitting can be employed to suppress the influence of outliers when performing STL decomposition. The key parameters for STL decomposition can be set as follows: the period parameter n_period can be set to an annual period, and a typical value for daily-scale data is 365; the seasonal smoothing window n_season is used to smooth interannual fluctuations in the same phase, and its value range is usually 7 to 365 days (should be an odd number), preferably 61 to 121 days, for example, 91 days; the trend smoothing window n_trend is used to extract the long-term trend term, and its value should be significantly larger than multiple annual periods to avoid absorbing interannual variability. Its value range is usually several years to a reasonable span of the covered sequence length, preferably about 10 years (about 3651 days for daily-scale data), and it must be an odd number; the low-pass filter window n_lowpass is used to separate the low-frequency components of the seasonal term in the iteration, and its value range is usually not less than the period length and should be an odd number, preferably about 2 years (about 731 days for daily-scale data).

[0023] After completing the STL decomposition, the long-term trend term can be removed and reconstructed using the following formula to obtain the detrended temperature series: AT_detrended(t)=AT_season(t)+AT_resid(t)+mean(AT_trend) , in, For long-term trend items The mean value within the analysis period is used to re-align the detrended series to the same average level as the original series. The resulting... The intra-annual and inter-annual variability of the original sequence was preserved, but multi-decadal warming signals were removed.

[0024] This implementation method achieves trend stripping through STL decomposition. While preserving the natural fluctuation characteristics of the temperature series within and between years, it independently separates the warming signals at multiple decadal scales. This enables the subsequent water temperature series reconstructed based on the detrended temperature series without long-term warming scenarios to accurately reflect water temperature processes driven only by intra-year climate variability, thereby improving the accuracy of distinguishing between the contribution of long-term climate warming and the contribution of intra-year climate variability in the attribution decomposition.

[0025] In some implementations, before decomposing the temperature time series, the process includes identifying and removing data from anomalous years: constructing a daily empirical distribution for the temperature time series by calendar day, determining an upper empirical quantile threshold and a lower empirical quantile threshold based on the daily empirical distribution; identifying calendar days where the observed temperature exceeds the upper empirical quantile threshold or is lower than the lower empirical quantile threshold as extreme days; calculating the proportion of extreme days to the number of valid days in each year, determining years where the proportion exceeds a preset anomalous year determination threshold as climate anomalous years, and removing the data corresponding to the climate anomalous years from the temperature time series.

[0026] Daily empirical distribution refers to collecting all temperature sample values ​​from multiple years of observation records for each calendar day, forming an empirical sample set for that calendar day. Based on this empirical sample set, its i-th value can be selected... Percentile and th Percentile is used as the upper empirical quantile threshold for that day. Compared with the lower empirical quantile threshold When the observed temperature on a certain day exceeds When the temperature is high, the day can be identified as an extreme high-temperature day; when it is lower than... When this occurs, the day can be identified as an extreme low temperature day. The value of the quantile threshold can be determined based on the actual data characteristics and analysis needs; for example, it can be taken as... , Specifically, the 95th and 5th percentiles are used as the thresholds for determining extreme high and low temperatures. After identifying extreme days, the proportion of extreme days to the total number of valid days in each year can be further calculated. When the proportion of extreme days in a given year exceeds the preset threshold for determining an abnormal year, that year is identified as a climate anomaly year, and its corresponding data is removed from the temperature time series to avoid interference from abnormal climate events on subsequent trend decomposition and model training results. The threshold for determining an abnormal year can be determined based on the climate characteristics and data quality of the study area.

[0027] This implementation method introduces an abnormal year screening step before trend stripping, which can reduce the interference of extreme climate events on long-term trend extraction and model training, making the trend decomposition results more stably reflect the normal variation pattern of temperature series, and improving the reliability of subsequent counterfactual reconstruction and attribution analysis.

[0028] Step S2: Using a pre-trained artificial intelligence time series model, the original temperature series and upstream hydrological features are used as model-driven inputs to reconstruct the water temperature series under the damless scenario; the detrended temperature series and upstream hydrological features are used as model-driven inputs to reconstruct the water temperature series under the long-term warming scenario.

[0029] An artificial intelligence time series model can be used to establish a mapping relationship between meteorological and hydrological driving conditions and downstream natural water temperature. Based on this mapping relationship, water temperature sequences for corresponding scenarios can be reconstructed under different climatic driving input conditions. The artificial intelligence time series model can be trained on samples from natural periods or periods of relatively weak regulation before the construction of the target reservoir, enabling it to learn the response relationship between meteorological and hydrological driving forces and downstream water temperature under natural conditions. In practical applications, when the sample size before the construction and operation of the target reservoir system is insufficient, periods with lower regulation intensity in the early stages of operation, such as the diversion period or trial operation period before reservoir impoundment, can be used as auxiliary training samples.

[0030] In one embodiment of this disclosure, the artificial intelligence time series model may employ a Long Short-Term Memory (LSTM) network model. LSTM is a gated recurrent neural network that, through the coordinated control of input gates, forget gates, and output gates, can selectively retain and transmit information over long time series, making it suitable for characterizing the long-lag nonlinear coupling relationship between meteorological and hydrological drivers and water temperature response. Model input may include a temperature series as a meteorological driver variable and upstream hydrological features as a hydrological driver variable. In a specific implementation, the meteorological driver variable may be a daily-scale temperature series, and the hydrological driver variable may be a daily-scale flow series from upstream control stations. To ensure the natural attributes of the counterfactual reconstruction, downstream local flow directly regulated by cascades is not used as input; instead, a more naturally representative upstream flow series can be used as the hydrological driver. If necessary, other feature variables that are not directly affected by the target cascade regulation and can represent natural driving conditions can also be included.

[0031] Before feeding the model-driven input into the artificial intelligence time series model, the various feature variables entering the model can be standardized to eliminate differences in units and improve the model training stability and convergence efficiency. Standardization can be achieved using the Z-score standardization method, which involves subtracting the mean of each feature variable from its training set value and then dividing by its standard deviation, so that each feature variable has zero mean and unit variance before being input into the model.

[0032] In some implementations, before feeding the model-driven input into the AI ​​time series model, a seasonal phase feature is constructed and used as additional input to the AI ​​time series model to incorporate intra-annual seasonal location information into the model during reconstruction: the date numbers of the year are encoded as a sinusoidal seasonal term, the expression of which is: , Where DOY represents the date's sequence number within the year. This refers to the phase offset parameter. The phase offset parameter... The determination method can be as follows: based on the temperature time series, calculate the multi-year average temperature according to calendar days to obtain the daily climatological temperature series, and take the date number corresponding to the minimum value in the daily climatological temperature series as the phase offset parameter. .

[0033] Seasonal phase features can provide the model with a periodic encoding of the time position within the year. Natural river water temperature is governed by annual solar radiation and climate seasonality, exhibiting a cyclical variation of cold winters and warm summers. Encoding the date number DOY (ranging from 1 to 365) as a single sine term allows this feature to reach its minimum value near the coldest day, its maximum value in summer, and then decrease as DOY progresses, consistent with the natural cold-warm cycle of annual air and river water temperature. A phase shift parameter is introduced. It can align the zero phase of the sine term with the coldest day of the year. For example, based on multi-year diurnal temperature observation sequences from meteorological stations, the multi-year average temperature can be calculated daily according to DOY to obtain the daily climatological temperature sequence. ;Pick The DOY corresponding to the minimum value is taken as the phase. ,Right now: .

[0034] This alignment operation allows sinDOY to be minimized on the coldest day of the year, thus synchronizing with the seasonal rhythm of water temperature.

[0035] This implementation method constructs seasonal phase features that are synchronized with the annual temperature cycle, enabling the model to directly perceive the seasonal position of the current time step during training and reconstruction. This reduces the burden of indirect inference of implicit seasonal information on the model and helps improve the accuracy of water temperature reconstruction under different seasonal conditions.

[0036] The artificial intelligence time series model can employ a multi-layered stacked LSTM network structure. For example, this structure can consist of several stacked LSTM layers (typically 2-4 layers, e.g., 3 layers) connected in series. Dropout layers can be used between adjacent LSTM layers for regularization. The output of the last LSTM layer can be connected to a fully connected linear layer (Dense layer) to output the predicted water temperature of the target downstream station at the next time step. If the model's input on day t is the feature sequence of the previous L days, then the model's training objective can be to minimize the predicted water temperature on day t+1. With observed water temperature The prediction error between them.

[0037] Model training can use mean squared error (MSE) as the loss function: , Where N is the number of samples in the training batch. The optimizer can be the Adam adaptive moment estimator. During training, early stopping and adaptive learning rate decay callback strategies can be further employed to suppress overfitting and improve generalization ability.

[0038] The model can be optimized using grid search for several key hyperparameters, and the optimal model configuration is determined based on the validation set performance. The key hyperparameters, their value ranges, and typical values ​​are shown in the table below: After model training, the original temperature sequence (including long-term warming trends) can be input into the trained model along with the upstream control station flow sequence and seasonal phase features to reconstruct the water temperature sequence under the dam-free scenario. Similarly, the detrended temperature sequence, along with the upstream control station flow sequence and seasonal phase features, can be input into the same model to reconstruct the water temperature sequence under the scenario without long-term warming. The dam-free scenario water temperature sequence represents the downstream natural water temperature process under the condition of "no dam construction," while the detrended temperature sequence only retains intra-annual and inter-annual climate variability and does not carry long-term warming signals. Therefore, the water temperature sequence under the scenario without long-term warming represents the downstream water temperature process under the condition of "no long-term climate warming," retaining only intra-annual and inter-annual climate variability. The only difference between the two scenarios is whether the temperature input includes a long-term warming trend; all other driving conditions and model structures are completely identical, thus ensuring the independent separability of the long-term climate warming contribution in subsequent attribution decomposition.

[0039] Without departing from the concept of this disclosure, the artificial intelligence time series model can also be replaced by other deep learning models with time-dependent representation capabilities, including but not limited to: Gated Recurrent Unit (GRU), Bidirectional Long Short-Term Memory (Bi-LSTM), variants of stacked LSTM (such as attention-based stacked LSTM), Transformer-type time series models (such as Informer, Autoformer, TemporalFusion Transformer, etc.), Temporal Convolutional Network (TCN), Sequence-to-Sequence (Seq2Seq) models, and hybrid models incorporating the above structures. All of these alternative models can accept multivariate time series inputs and characterize the nonlinear and hysteretic effects of the input variables on the target water temperature, and can be selected according to specific application scenarios and data conditions.

[0040] Step S3: The actual observed water temperature sequence of the downstream control station, the water temperature sequence under the damless scenario, and the water temperature sequence under the long-term warming scenario are combined to form a three-scenario water temperature sequence system. Based on the statistical difference relationship between the sequences in the three-scenario water temperature sequence system, a multi-factor attribution decomposition is performed to decompose the total water temperature change into anthropogenic regulation contribution, long-term climate warming contribution, and annual climate variability contribution.

[0041] The actual observed water temperature sequence, i.e., the water temperature sequence actually observed by the downstream control station after the dam construction, is denoted as... The water temperature sequence under the no-dam scenario, namely the downstream natural water temperature sequence under the condition of "if no dam is built," is reconstructed using a trained artificial intelligence model based on the original air temperature (including long-term warming trends) and upstream hydrological drivers. It is denoted as... The downstream water temperature series under the "no long-term warming scenario" is a series reconstructed using the same artificial intelligence model based on detrended air temperature and upstream hydrological drivers, denoted as... The above three scenario sequences constitute a three-scenario water temperature sequence system.

[0042] The logical relationships between the three scenario sequences are as follows. and The comparison reveals the impact of human intervention on downstream water temperature, reflecting the combined effect of reservoir operation (including storage and release, thermal stratification, and changes in water intake elevation) on downstream water temperature under the background of long-term climate warming. and The comparison reveals an independent contribution from long-term climate warming, which reflects the sustained increase in downstream water temperature under natural conditions caused by multi-decadal-scale climate warming. After stripping away the long-term warming trend, only the natural water temperature process corresponding to the intra-annual and inter-annual climate variability is retained. The deviation of this process from the natural water temperature climatology of the same period before the dam construction can separate the contribution of intra-annual climate variability. This contribution reflects the natural modulation effect of meteorological conditions on downstream water temperature by fluctuations on seasonal and inter-annual scales.

[0043] The attribution decomposition in this step involves the total change in water temperature. Decomposed into contributions from human regulation Climate-driven contributions and residuals Climate-driven contributions This can be further broken down into long-term contributions to global warming. Contribution to intra-annual climate variability The decomposition relationship is as follows: , , Where Δ is positive, it indicates that the downstream water temperature has increased relative to the reference or counterfactual scenario; a negative value indicates that it has decreased. Residual term These contributions reflect biases that the model fails to explain and can be used to assess the reliability of the attribution results. Each contribution can be calculated daily on a daily scale. The results can then be aggregated into monthly, seasonal, or annual scale evaluation results as needed; alternatively, they can be calculated directly based on the mean difference of the corresponding time window.

[0044] In some implementations, the anthropogenic regulation contribution item includes an anthropogenic regulation contribution sequence and an anthropogenic regulation contribution degree; the long-term climate warming contribution item includes a long-term climate warming contribution sequence and a long-term climate warming contribution degree; and the intra-annual climate variability contribution item includes an intra-annual climate variability contribution sequence and an intra-annual climate variability contribution degree. Based on the regulation nodes of the target reservoir system, the time interval corresponding to the temperature time series is divided into a pre-regulation historical period and a post-regulation operating period. The multi-factor attribution decomposition based on the statistical difference relationship between the sequences in the three-scenario water temperature sequence system includes: for the post-regulation operating period, calculating the difference sequence between the actual observed water temperature sequence and the damless scenario water temperature sequence at the corresponding time step, as the anthropogenic regulation contribution sequence, and placing the anthropogenic regulation contribution sequence in the... The mean value during the post-regulation operating period is used as the contribution of human regulation. For the post-regulation operating period and the historical period before regulation, the statistical mean of the water temperature series under the no-dam scenario is calculated respectively, and the difference between the two is used as the climate-driven contribution, and the corresponding difference series is used as the climate-driven contribution series. For the post-regulation operating period and the historical period before regulation, the statistical mean of the water temperature series under the no-long-term warming scenario is calculated respectively, and the difference between the two is used as the contribution of intra-year climate variability, and the corresponding difference series is used as the intra-year climate variability contribution series. The difference between the climate-driven contribution series and the intra-year climate variability contribution series is calculated as the long-term climate warming contribution series, and the difference between the climate-driven contribution and the intra-year climate variability contribution is calculated as the long-term climate warming contribution.

[0045] In this implementation, the division between the historical period before regulation and the operational period after regulation is based on the actual construction and operation milestones of the target reservoir system. The regulation milestones correspond to the time nodes when the target reservoir system was completed and put into operation. Each component is defined by the difference of the statistical means of water temperature sequences under different scenarios using the following formula: , , , , , , in, This is a sequence of actual observed water temperatures; The natural water temperature sequence is reconstructed using an artificial intelligence model based on the original air temperature (including long-term warming trends), which is the water temperature sequence under the damless scenario in step S2; The water temperature sequence is reconstructed based on detrended air temperature using the same artificial intelligence model, namely the water temperature sequence without long-term warming scenario in step S2.

[0046] Human regulation contribution Used to characterize the impact of reservoir operation, storage and release scheduling, and related engineering controls on downstream water temperature. Long-term contribution to global warming. Used to characterize the sustained impact of multi-decade warming on downstream water temperature. Contribution of intra-annual climate variability. It is used to characterize the natural modulation effect caused by the fluctuations in meteorological conditions on seasonal and interannual scales.

[0047] This implementation method achieves independent quantitative separation of the three contributions—human intervention, long-term climate warming, and intra-annual climate variability—by dividing the time periods before and after regulation and calculating each attribution component item by the difference in the statistical means of the three scenario sequences within each time period. Specifically, the contribution of long-term climate warming is indirectly obtained through the difference between the climate-driven contribution and the intra-annual climate variability contribution, avoiding the uncertainty introduced by trend extraction errors when directly estimating the long-term trend impact.

[0048] In some implementations, after completing the multi-factor attribution decomposition, the method further includes identifying dominant driving factors: calculating the percentage of the absolute values ​​of the contributions of human regulation, long-term climate warming, and intra-annual climate variability to the sum of the absolute values ​​of all contributions, as relative contribution rates; identifying the attribution factors corresponding to contributions whose relative contribution rates are greater than or equal to a preset single-factor dominant threshold as dominant driving factors; and identifying the attribution factors corresponding to multiple contributions whose relative contribution rates are all greater than or equal to a preset joint dominant threshold as joint dominant driving factors, wherein the attribution factors include human regulation, long-term climate warming, and intra-annual climate variability.

[0049] The relative contribution rate is defined as the percentage of the absolute value of each component to the sum of the absolute values ​​of all components: , Where i represents a certain attribution component (such as ANT, Trend, NCV, etc.). This is the sum of the absolute values ​​of all components participating in the normalization. Using a normalization method based on absolute value proportions can simultaneously reflect the contribution intensity of each component in the heating and cooling directions, avoiding the underestimation of the dominant factor due to positive and negative cancellation. When determining the dominant driving factor, if the relative contribution rate of a component exceeds the single-factor dominance threshold, that component can be identified as the dominant driving factor; for example, the single-factor dominance threshold can be set to 60%. If the relative contribution rates of two or more components all exceed the joint dominance threshold, these components can be identified as the joint dominant driving factor; for example, the joint dominance threshold can be set to 40%.

[0050] This implementation method introduces relative contribution rate and dominant factor identification rules based on the attribution decomposition results, so that the evaluation conclusions not only include the absolute contribution of each factor, but also quantitatively determine which factor is dominant in a specific period or time scale, providing a clearer attribution basis for subsequent decisions on reservoir scheduling optimization and thermal risk management.

[0051] Step S4: Based on the results of the multi-factor attribution decomposition, perform cumulative effect calculation on the water temperature process during the target evaluation period to obtain the cross-seasonal heat accumulation diagnosis results.

[0052] Based on the attribution results, cumulative effect calculations can be performed on water temperature processes within a specific evaluation period, transforming diurnal attribution information into diagnostic results that reflect cross-seasonal heat accumulation characteristics. The target evaluation period can be determined according to evaluation needs, such as the autumn water storage period (e.g., September to November), the winter period (e.g., December to February of the following year), or other periods of interest.

[0053] In some implementations, obtaining the cross-seasonal heat accumulation diagnostic results includes at least one of the following: summing the anthropogenic regulation contribution sequence over the target evaluation period to obtain a regulation-induced heat index. To quantitatively assess the intensity of cumulative heat exposure induced by reservoir regulation during this period: Where Ω represents the target evaluation period. The value at time step t in the artificially regulated contribution sequence is used; the actual observed water temperature sequence is cumulatively summed over the target evaluation period to obtain the observed heat index. ,in, Let t be the value of the t-th time step in the actual observed water temperature sequence.

[0054] The above calorie indicators and (Hereinafter referred to as Category D indicators) all have the dimension of ℃·day. Induced heat regulation indicators By cumulatively summing the contribution sequence of human regulation over the target evaluation period, the intensity of cumulative warming or cooling caused by reservoir regulation during that period can be quantitatively described. (Observation of heat indicators) By summing the actual observed water temperature series over the same time period, the cumulative heat level experienced by the downstream river section during that period can be quantitatively described. Combining these two methods can extend the attribution decomposition results from daily-scale bias to time-scale heat accumulation, thereby revealing the heat exposure effect of reservoir regulation within a specific time period.

[0055] This implementation method transforms the daily-scale attribution information into a heat exposure index with clear physical dimensions by summing the attribution sequence and the observation sequence over time periods. This allows the evaluation results to reflect the intensity of the cumulative heat effect of reservoir regulation during the target period, rather than being limited to deviation descriptions at the daily or seasonal average levels.

[0056] In some embodiments, obtaining the cross-seasonal heat accumulation diagnostic results further includes performing water storage time series characteristic diagnosis and / or performing warm water storage timing diagnosis; wherein, performing water storage time series characteristic diagnosis includes: calculating the net storage process of the target reservoir system during the target evaluation period. : , in, Let be the inflow rate of the k-th reservoir at time step t. Let K be the outflow from the k-th reservoir at time step t, and K be the total number of reservoirs in the target reservoir system; calculate the water storage time series index based on the net storage process. : , Wherein, τ(t) is the normalized time index within the target evaluation period, and its value ranges from [0,1]. The normalized first-order moment of time, weighted by the net accumulation process; The diagnosis of warm water storage timing includes: calculating a warm water storage timing index based on the following formula to characterize the coupling relationship between storage time and storage temperature. : , , in, Let be the inflow water temperature observation value at the t-th time step, i.e., the water temperature observation value at the inflow section of the target reservoir system. The normalized first moment of time is the product of the net storage process and the inflow water temperature as the joint weight.

[0057] Net storage process Defined as the difference between the daily inflow and daily outflow of the target reservoir system, with negative values ​​truncated to retain only the net inflow during the impoundment phase. K is set to 1 when the target reservoir system is a single reservoir, and K is set to the corresponding total number of reservoirs when it is a cascade system. In the case that there is no significant interval of inflow during the impoundment phase of a cascade system... It can characterize the daily net water storage at the system level.

[0058] Water storage time series indicators Time centroid based on the net accumulation process Calculated. For The normalized time first moment of the weights reflects the centroid position of net reserves within the target evaluation period. The normalized time index τ(t) takes a linear distribution from 0 (start of the period) to 1 (end of the period) within the target evaluation period. The larger the value, the earlier the water storage begins, meaning the focus of the water storage period is closer to the beginning of the period. The smaller the value, the later the water storage period begins, meaning the focus of the water storage period is closer to the end of the period.

[0059] Warm water storage timing indicators Based on the net storage process, the inflow water temperature is further introduced. As a joint weight. Compared with net reserves. different, Simultaneously consider both water storage time and water temperature. When the water temperature is high and the net storage volume is large in the early stages of the water storage phase... A larger value can be selected, corresponding to the warm water pre-storage mode; when large amounts of water are stored only in the later stages of the water storage phase, or when only colder water is stored, A smaller value can be selected, corresponding to the mode of post-cooled water storage.

[0060] The above-mentioned Category D heat indicators and water storage time series indicators Warm water storage timing indicators These are all indicators for cross-seasonal heat accumulation diagnosis, and they can collectively constitute a cross-seasonal heat accumulation diagnostic indicator system. Category D indicators reflect the intensity of heat accumulation. Reflecting the temporal distribution characteristics of water storage, This further demonstrates the coupling relationship between heat and time. This implementation method introduces a time dimension and a heat-time coupling dimension diagnosis on the basis of heat indicators, so that the evaluation results can reveal the intrinsic relationship between the water storage sequence arrangement and the downstream heat accumulation effect, rather than being limited to a single dimension description of heat accumulation intensity.

[0061] In some implementations, the method further includes: establishing a statistical correlation between the water storage time series index and / or the warm water storage timing index and the heat index in the cross-seasonal heat accumulation diagnostic results; and generating a causal diagnostic conclusion between the water storage time series and the downstream warming intensity based on the statistical correlation.

[0062] There are no particular limitations on how statistical associations are established; correlation analysis, regression analysis, or other methods that can describe the strength of the statistical association or functional relationship between the independent and dependent variables can be used. For example, calculations can be performed... or and The Pearson correlation coefficient or Spearman rank correlation coefficient between the two factors can be used to assess the statistical reliability of the association through significance testing; alternatively, a correlation coefficient can be established. or As the independent variable, with The regression model, with the dependent variable as the dependent variable, quantitatively describes the direction and extent of the influence of water storage timing characteristics on downstream heat accumulation intensity. Through the aforementioned statistical correlation analysis, causal diagnostic conclusions can be further generated, such as determining whether there is a statistically positive correlation between pre-autumn water storage and increased downstream warming intensity in winter, thus providing quantitative basis for reservoir scheduling optimization and thermal risk management.

[0063] This implementation method further establishes statistical correlations between indicators based on the construction of diagnostic indicators, thereby elevating the evaluation conclusions from descriptive statistics of a single indicator to correlation analysis and causal diagnosis between indicators, which helps to identify the response patterns between water storage scheduling strategies and downstream heat accumulation effects.

[0064] Please refer to Figure 2 , Figure 2 It shows a comparison of river water temperature before and after the reservoir construction. For example... Figure 2 As shown on the far left, before the dam was built, the river was in a free-flowing state. (As shown...) Figure 2 As shown in the two dam cross-sections on the right, after the dam was built, the reservoir had a lower water level during the warm season, and a stable temperature stratification structure could be formed in the reservoir area; during the cold season, the water level was higher, and the temperature stratification in the reservoir area was weaker, demonstrating the impact of reservoir storage and release scheduling and seasonal water level changes on the thermal stratification structure of the reservoir area.

[0065] Figure 3 This is a schematic diagram of the technical route for an artificial intelligence-based method for assessing the cumulative impact of reservoir water temperature, provided in an exemplary embodiment of this disclosure. Regarding time period division, based on the construction and operation progress of the target reservoir system, the analysis period is divided into the pre-dam period (2000–2011), the construction transition period (2012–2014), and the post-dam operation period (2015–2020). Samples from the pre-dam period can be used for training and validation of the artificial intelligence time series model, while samples from the post-dam operation period can be used for counterfactual reconstruction and attribution decomposition. The construction transition period is the stage where the reservoir is gradually built and impounded. During this period, the downstream water temperature is simultaneously affected by construction disturbances and initial impoundment, and the regulation characteristics are still unstable. Samples from this period can be selected for inclusion or exclusion based on actual conditions during model training and attribution decomposition.

[0066] Figure 3The model's input variables and training process were also demonstrated. Input variables include the air temperature at Zhutuo station (Zhutuo AT), seasonal phase characteristics (SinDOY), and the flow rate at Huatan station (Huatan Q). After Z-score standardization, these variables are fed into the LSTM model, with the training target being the water temperature at Zhutuo station (Zhutuo WT). The model optimizes hyperparameters and input sequence length through grid search, and selects the optimal model configuration on the validation set based on three evaluation metrics: Nash-Sutcliffe Efficiency (NSE), Root Mean Square Error (RMSE), and Mean Absolute Error (MAE). Using these three metrics together allows for a comprehensive evaluation of model performance from three perspectives: goodness of fit, error magnitude, and error robustness.

[0067] Figure 3 It also demonstrates the difference relationship between the three scenario water temperature sequences. Actual water temperature observed after dam construction. Compared with the actual water temperature observed before dam construction The difference between them corresponds to the total water temperature change ΔTOT; the actual water temperature observed after dam construction. Simulated natural water temperature after dam construction The difference between them corresponds to the contribution of human intervention ΔANT; the simulated natural water temperature after dam construction Simulated natural water temperature before dam construction The difference between them corresponds to the climate driving contribution ΔCLI, which can be further decomposed into the long-term climate warming contribution ΔTrend and the intra-annual climate variability contribution ΔNCV. The figure also shows the process of decomposing the temperature series into a seasonal, residual, and trend term using STL, corresponding to the trend stripping operation in this scheme.

[0068] like Figure 3As shown, the attribution results are summarized into two main categories: Climate Change and Anthropogenic Regulation. Climate Change can be further divided into Long-Term Climate Change and Natural Climate Variability. Anthropogenic Regulation can encompass the water temperature impacts under different regulation modes, such as Ecological Regulation, Routine Operation, and Sediment-flushing. The above classification corresponds to the multi-factor attribution decomposition in this scheme, where the two sub-items of Climate Change correspond to ΔTrend and ΔNCV, respectively, and Anthropogenic Regulation corresponds to ΔANT.

[0069] This disclosure combines artificial intelligence time series modeling, long-term temperature trend stripping, multi-scenario counterfactual reconstruction, and multi-factor attribution decomposition to achieve independent quantitative attribution of downstream water temperature changes by three factors: reservoir regulation, long-term climate warming, and intra-annual climate variability, even in the absence of concurrent natural observational controls. Based on the attribution results, cross-seasonal heat accumulation diagnostic indicators such as heat exposure, water storage sequence, and warm water storage timing are further constructed, enabling the evaluation results to cover multiple time scales, including daily, monthly, seasonal, and interannual scales. The method described in this disclosure can be used for historical attribution analysis as well as for comparative evaluation between different operational schemes, providing quantitative technical support for reservoir scheduling optimization and downstream thermal ecological risk prevention and control.

[0070] Figure 4 An electronic device provided in at least one embodiment of the present disclosure includes a memory and a processor. The memory is used to store computer instructions that can be executed on the processor. The processor is used to implement the artificial intelligence-based reservoir water temperature cumulative impact assessment method according to any embodiment or implementation of the present disclosure when executing the computer instructions.

[0071] At least one embodiment of this disclosure also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the artificial intelligence-based method for assessing the cumulative impact of reservoir water temperature as described in any embodiment or implementation of this disclosure.

[0072] Those skilled in the art will understand that one or more embodiments of this specification can be provided as a method, system, or computer program product. Therefore, one or more embodiments of this specification may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, one or more embodiments of this specification may take 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.

[0073] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the data processing device embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0074] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0075] The embodiments of the subject matter and functional operation described in this specification can be implemented in: digital electronic circuits, tangibly embodied computer software or firmware, computer hardware including the structures disclosed in this specification and their structural equivalents, or combinations thereof. Embodiments of the subject matter described in this specification can be implemented as one or more computer programs, i.e., one or more modules of computer program instructions encoded on a tangible, non-transitory program carrier for execution by a data processing apparatus or for controlling the operation of a data processing apparatus. Alternatively or additionally, the program instructions may be encoded on artificially generated propagation signals, such as machine-generated electrical, optical, or electromagnetic signals, which are generated to encode information and transmit it to a suitable receiving device for execution by the data processing apparatus. The computer storage medium may be a machine-readable storage device, a machine-readable storage substrate, a random or serial access memory device, or combinations thereof.

[0076] The processing and logic flow described in this specification can be executed by one or more programmable computers that execute one or more computer programs to perform corresponding functions by operating on input data and generating output. The processing and logic flow can also be executed by dedicated logic circuitry—such as FPGAs (Field-Programmable Gate Arrays) or ASICs (Application-Specific Integrated Circuits), and the device can also be implemented as dedicated logic circuitry.

[0077] Suitable computers for executing computer programs include, for example, general-purpose and / or special-purpose microprocessors, or any other type of central processing unit. Typically, the central processing unit receives instructions and data from read-only memory and / or random access memory. The basic components of a computer include a central processing unit for implementing or executing instructions and one or more memory devices for storing instructions and data. Typically, a computer will also include one or more mass storage devices for storing data, such as disks, magneto-optical disks, or optical disks, or the computer will be operatively coupled to such mass storage devices to receive data from or transfer data to them, or both. However, a computer is not required to have such devices. Furthermore, a computer can be embedded in another device, such as a mobile phone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a global positioning system (GPS) receiver, or a portable storage device such as a universal serial bus (USB) flash drive, to name a few.

[0078] Computer-readable media suitable for storing computer program instructions and data include all forms of non-volatile memory, media, and memory devices, such as semiconductor memory devices (e.g., EPROM, EEPROM, and flash memory devices), magnetic disks (e.g., internal hard disks or removable disks), magneto-optical disks, and CD-ROM and DVD-ROM disks. Processors and memory may be supplemented by or incorporated into dedicated logic circuitry.

[0079] While this specification contains numerous specific implementation details, these should not be construed as limiting the scope of any invention or the scope of the claims, but rather are primarily intended to describe features of specific embodiments of a particular invention. Certain features described in the various embodiments herein may also be implemented in combination in a single embodiment. Conversely, various features described in a single embodiment may also be implemented separately in various embodiments or in any suitable sub-combination. Furthermore, while features may function in certain combinations as described above and even initially claimed in this way, one or more features from a claimed combination may be removed from that combination in some cases, and a claimed combination may refer to a sub-combination or a variation thereof.

[0080] Similarly, although the operations are depicted in a specific order in the accompanying drawings, this should not be construed as requiring these operations to be performed in the specific order shown or sequentially, or requiring all illustrated operations to be performed to achieve the desired result. In some cases, multitasking and parallel processing may be advantageous. Furthermore, the separation of various system modules and components in the above embodiments should not be construed as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product or packaged into multiple software products.

[0081] Thus, specific embodiments of the subject matter have been described. Other embodiments are within the scope of the appended claims. In some cases, the actions recited in the claims may be performed in a different order and still achieve the desired result. Furthermore, the processes depicted in the drawings are not necessarily shown in a specific order or sequence to achieve the desired result. In some implementations, multitasking and parallel processing may be advantageous.

[0082] The above description is merely a preferred embodiment of one or more embodiments of this specification and is not intended to limit the scope of one or more embodiments of this specification. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments of this specification should be included within the protection scope of one or more embodiments of this specification.

Claims

1. A method for assessing the cumulative impact of reservoir water temperature based on artificial intelligence, characterized in that, include: The temperature time series of the river section where the target reservoir system is located is decomposed and the long-term warming trend term is removed to obtain the detrended temperature series; Using a pre-trained artificial intelligence time series model, the original temperature series and upstream hydrological features are used as model-driven inputs to reconstruct the water temperature series under the dam-free scenario, and the detrended temperature series and upstream hydrological features are used as model-driven inputs to reconstruct the water temperature series under the long-term warming scenario. The actual observed water temperature sequence of the downstream control station, the water temperature sequence under the damless scenario, and the water temperature sequence under the long-term warming scenario are used to form a three-scenario water temperature sequence system. Based on the statistical difference relationship between the sequences in the three-scenario water temperature sequence system, a multi-factor attribution decomposition is performed to decompose the total water temperature change into anthropogenic regulation contribution, long-term climate warming contribution, and annual climate variability contribution. Based on the results of the multi-factor attribution decomposition, the cumulative effect calculation of the water temperature process during the target evaluation period is performed to obtain the cross-seasonal heat accumulation diagnosis results.

2. The method according to claim 1, characterized in that, The process of decomposing the temperature time series and removing the long-term warming trend term includes: The temperature time series is decomposed into a long-term trend term, a seasonal term, and a residual term using a locally weighted regression-based seasonal decomposition method. The long-term trend term is removed from the temperature time series, and the mean of the long-term trend term during the analysis period is added back to obtain the detrended temperature series.

3. The method according to claim 1, characterized in that, Before decomposing the temperature time series, the process also includes identifying and removing abnormal years: A daily empirical distribution is constructed for the temperature time series according to the calendar days, and the upper empirical quantile threshold and the lower empirical quantile threshold are determined based on the daily empirical distribution; Calendar days in which the observed temperature exceeds the upper empirical quantile threshold or falls below the lower empirical quantile threshold are identified as extreme days. The proportion of extreme days to the number of valid days in each year is statistically analyzed. Years with a proportion exceeding a preset threshold for determining abnormal years are identified as climate abnormal years, and the data corresponding to the climate abnormal years are removed from the temperature time series.

4. The method according to claim 1, characterized in that, Before feeding the model-driven input into the AI ​​time series model, the method further includes constructing seasonal phase features and using them as additional input to the AI ​​time series model, so that the model incorporates intra-annual seasonal location information during the reconstruction process. The dates of the year are encoded as sinusoidal seasonal terms, the expression of which is: ， Where DOY represents the date's sequence number within the year. This refers to the phase offset parameter; The phase offset parameter The determination method is as follows: based on the temperature time series, the multi-year average temperature is calculated according to calendar days to obtain the daily climatological temperature series, and the date number corresponding to the minimum value in the daily climatological temperature series is used as the phase offset parameter. .

5. The method according to claim 1, characterized in that, The anthropogenic regulation contribution item includes an anthropogenic regulation contribution sequence and anthropogenic regulation contribution degree; the long-term climate warming contribution item includes a long-term climate warming contribution sequence and a long-term climate warming contribution degree; and the intra-annual climate variability contribution item includes an intra-annual climate variability contribution sequence and an intra-annual climate variability contribution degree. Based on the control nodes of the target reservoir system, the time interval corresponding to the temperature time series is divided into the historical period before control and the operating period after control. The multi-factor attribution decomposition based on the statistical difference relationship between the sequences in the three-scenario water temperature sequence system includes: For the post-regulation operation period, the difference sequence between the actual observed water temperature sequence and the water temperature sequence under the damless scenario is calculated at the corresponding time step, which is used as the human-controlled contribution sequence, and the mean of the human-controlled contribution sequence during the post-regulation operation period is used as the human-controlled contribution degree. For the post-regulation operation period and the pre-regulation historical period, the statistical mean of the water temperature sequence under the damless scenario is calculated respectively. The difference between the two is taken as the climate driving contribution, and the corresponding difference sequence is taken as the climate driving contribution sequence. For the post-regulation operating period and the pre-regulation historical period, the statistical mean of the water temperature sequence without long-term warming scenario is calculated respectively. The difference between the two is taken as the contribution of the annual climate variability, and the corresponding difference sequence is taken as the contribution sequence of the annual climate variability. The difference between the climate-driven contribution sequence and the intra-annual climate variability contribution sequence is calculated as the long-term climate warming contribution sequence, and the difference between the climate-driven contribution degree and the intra-annual climate variability contribution degree is calculated as the long-term climate warming contribution degree.

6. The method according to claim 5, characterized in that, After completing the multi-factor attribution decomposition, the process also includes identifying the dominant driving factors: The absolute values ​​of the contributions of human regulation, long-term climate warming, and annual climate variability are calculated as percentages of the sum of the absolute values ​​of all contributions, and are used as relative contribution rates. The attribution factors corresponding to the relative contribution rates that are greater than or equal to the preset single-factor dominance threshold are identified as dominant driving factors. The attribution factors corresponding to multiple contributions whose relative contribution rates are all greater than or equal to the preset joint dominance threshold are identified as joint dominant driving factors. The attribution factors include human regulation, long-term climate warming, and intra-annual climate variability.

7. The method according to claim 5, characterized in that, Obtaining the cross-seasonal heat accumulation diagnostic results includes at least one of the following: The cumulative summation of the artificially induced contribution sequence over the target evaluation period yields the induced heat index. To quantitatively assess the intensity of cumulative heat exposure induced by reservoir regulation during this period: ， in, For the target evaluation period, The value of the t-th time step in the artificially regulated contribution sequence; The actual observed water temperature series is cumulatively summed over the target evaluation period to obtain the observed heat index. : ， in, Let t be the value of the t-th time step in the actual observed water temperature sequence.

8. The method according to claim 7, characterized in that, Obtaining the cross-seasonal heat accumulation diagnostic results also includes performing water storage time sequence characteristic diagnosis and / or warm water storage timing diagnosis; Among them, the diagnosis of water storage time-series characteristics includes: Calculate the net storage process of the target reservoir system during the target evaluation period. : , in, Let be the inflow rate of the k-th reservoir at time step t. Let K be the outflow from the k-th reservoir at time step t, and K be the total number of reservoirs in the target reservoir system. Calculate water storage time series indicators based on the net storage process. : ， ， Wherein, τ(t) is the normalized time index within the target evaluation period, and its value ranges from [0,1]. The normalized first-order moment of time, weighted by the net accumulation process; Diagnostic procedures for warm water storage include: The warm water storage timing index, used to characterize the coupling relationship between storage time and storage temperature, is calculated based on the following formula. : ， ， in, Let be the inflow water temperature observation value at time step t. The normalized first moment of time is the product of the net storage process and the inflow water temperature as the joint weight.

9. The method according to claim 8, characterized in that, Also includes: Establish a statistical correlation between the water storage time series index and / or the warm water storage timing index and the heat index in the cross-seasonal heat accumulation diagnosis results; Based on the statistical correlation, a causal diagnostic conclusion is generated regarding the relationship between water storage time sequence and downstream warming intensity.

10. An electronic device, characterized in that, The device includes a memory and a processor, the memory being used to store computer instructions executable on the processor, and the processor being used to implement the method of any one of claims 1 to 9 when executing the computer instructions.

Images