A water quality index interval prediction method under non-consistent scenarios

Abstract

The present invention discloses a method for interval prediction of water quality indicators under non-consistent scenarios, including the following steps: obtaining a runoff time series and a water quality time series within a specified time period, and performing a comprehensive judgment of non-consistency; using a variety of probability distribution models, respectively fitting the runoff time series and the water quality time series, and determining the best probability distribution model; constructing an AR model to realize random simulation of runoff, and output a simulated runoff sequence; adjusting the simulated runoff sequence to obtain a non-consistent scenario; constructing a two-dimensional joint Copula function model, inputting future non-consistent scenarios into the two-dimensional joint Copula function model, and outputting a water quality prediction sequence set; extracting a residual sample set from a residual set between the simulated value and the measured value of a water quality point in the actual measurement period from the simulated sequence set of water quality, and realizing interval prediction of water quality indicators according to the residual sample set. According to the above technical scheme, point prediction of water quality data and water quality interval prediction under the background of non-consistency can be obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of water quality index prediction, and in particular to a method for water quality index interval prediction in an inconsistent scenario. Background Art

[0002] With the development of society, the consumption of domestic water, industrial water and agricultural water is increasing, and the demand for water resources is also increasing. However, the increasingly severe water pollution has become one of the main water problems facing my country. Under this circumstance, the simulation and prediction of various water quality indicators that characterize the water pollution status has also become a hot spot and key in water quality research.

[0003] In actual use, various indicators (such as pH value, potassium permanganate index, ammonia nitrogen, total phosphorus, total nitrogen, etc.) are often monitored by collecting water samples, and water bodies are divided into five categories according to the limit values ​​of various water quality indicators in the "Surface Water Environmental Quality Standards".

[0004] At present, many studies have been conducted on water quality simulation. Most of these studies use statistical models, hydrological models, machine learning and deep learning, as well as the coupling of multiple methods, to simulate water quality based on other hydrological data series in the historical period, and then derive the corresponding water quality index series. However, under the changing environment, the hydrological inconsistency process on a global scale is becoming more and more obvious, and the traditional hydrological consistency is no longer applicable to the actual environment. Therefore, the rationality of the water quality simulation results based on historical hydrological data needs to be further confirmed, and when the hydrological data show inconsistency, it is necessary to further analyze and simulate the water quality index data that are correlated with the hydrological data.

[0005] The characteristics of inconsistent hydrological data are that its statistical characteristics and parameters change over time, which makes the traditional view that the hydrological simulation process is "steady state" no longer applicable. The commonly used scheme is to make point predictions of water quality based on the consistency assumption, that is, the input sequence does not consider the statistical characteristics and parameters that change over time, and cannot highlight the uncertainty of water quality simulation data. Therefore, a water quality simulation method is needed that starts from the background of inconsistency, comprehensively considers the uncertainty in multiple inconsistent inputs and the representativeness of water quality samples, and realizes point prediction and interval prediction of inconsistent water quality data. Summary of the invention

[0006] To achieve the above purpose, the present application provides a method for water quality index interval prediction in an inconsistent scenario, comprising the following steps:

[0007] Obtain the runoff time series and water quality time series during the measurement period and perform comprehensive judgment on inconsistency;

[0008] A variety of probability distribution models are used to fit the runoff time series and water quality time series, and the best probability distribution model is determined based on the goodness of fit test.

[0009] Construct an AR model to realize random simulation of runoff and output the predicted simulated runoff sequence; adjust the simulated runoff sequence to obtain inconsistent scenarios;

[0010] Construct a two-dimensional joint Copula function model, input non-consistent scenarios into the two-dimensional joint Copula function model, and output a water quality prediction sequence set;

[0011] The measured runoff during the measurement period is obtained, and the measured runoff is input into the two-dimensional joint Copula function model to obtain a simulation sequence set of water quality, and a residual set between the simulated value and the measured value of the water quality point during the measurement period is extracted to extract a residual sample set, and the water quality index interval prediction is realized according to the residual sample set.

[0012] Among them, executing the comprehensive judgment of inconsistency includes:

[0013] The MK trend analysis method is used to obtain the changing trend of each time series, including: using the standard normal distribution for hypothesis testing, and judging the trend change of the time series by the critical value of the MK statistic being greater than or less than the 95% confidence level;

[0014] The inconsistency background of each time series is determined by the MK mutation test method, including: taking a 95% confidence level, determining whether the test quantity curve and the confidence interval intersect. If they intersect, it is determined that the time series has a significant mutation at the intersection, showing inconsistency;

[0015] The correlation degree between the runoff sequence and the water quality sequence is obtained through correlation analysis, including: when the correlation coefficient is greater than 0, it means that there is a positive correlation between the two sequences, otherwise there is a negative correlation; the closer its absolute value is to 1, the greater the degree of correlation.

[0016] Further, using a plurality of probability distribution models includes: determining probability distribution characteristics of the runoff time series and each water quality time series, the probability distribution characteristics including: extreme value distribution, exponential distribution, gamma distribution, generalized extreme value distribution, generalized Pareto distribution, lognormal distribution, normal distribution and Weibull distribution;

[0017] Goodness of fit includes: AIC value, RMSE value and KS test statistics;

[0018] The basis for determining the best probability distribution model based on the goodness of fit test includes: when the statistic of the KS test is equal to 1, it means that the distribution characteristics meet the hypothesis; the smaller the RMSE value, the better the fit; the smaller the AIC value, the better the fit.

[0019] The AR model consists of the random disturbance term The weighted sum of some items of the time series is composed of:

[0020] ,in, is the simulated runoff value at time t, is the autoregressive order, is the autoregression coefficient, is the noise term;

[0021] Adjusting the simulated runoff sequence refers to simulating inconsistent scenarios through statistical factor adjustment and seasonal variation of runoff; the inconsistent scenarios include: mean increase, mean decrease, rise, Decline, season advance, season lag;

[0022] Furthermore, the statistical factor adjustment method is used to simulate inconsistent scenarios according to the impact of inconsistent conditions on the statistical characteristic parameters of the sequence, including mean inconsistency and Inconsistency;

[0023] When generating the mean scenario, ,in, and They are the runoff sequence under the mean inconsistency scenario and the simulated runoff sequence obtained by the AR model, is the adjustment factor;

[0024] generate In the scenario of , while ensuring that the mean remains unchanged, ,in, express Runoff sequences under non-consistent scenarios; is the mean runoff value.

[0025] Furthermore, seasonal variation of runoff is used to reflect the impact of non-uniform conditions on runoff seasonality;

[0026] The seasonal variation of runoff is realized by the cumulative time distribution function to reconstruct the runoff under seasonal inconsistency conditions. The reconstruction model is expressed as:

[0027] ,

[0028] ,

[0029] ,

[0030] in, For the time series The cumulative time distribution function, is the length of the time series, and They represent the measured runoff time series and the runoff time series under the seasonal inconsistency scenario, respectively. , and They represent the cumulative time distribution functions of the triangular inflow process lines under the inconsistency scenario, before the inconsistency change, and after the inconsistency change, respectively.

[0031] Building a two-dimensional joint Copula function model includes:

[0032] Determine the water quality index sequence used in the simulation ;run-off ;

[0033] Determine marginal distributions of runoff and water quality indicator series;

[0034] Select an appropriate Copula function to describe the correlation structure between runoff and water quality indicators.

[0035] Furthermore, when extracting the residual sample set, the Bootstrap method is used to extract the residual set between the simulated value and the measured value of the water quality point in the actual measurement period with replacement. times, get A new set of residual samples.

[0036] Realizing interval prediction of water quality indicators based on the residual sample set means: performing interval prediction of simulated residuals at 95% and 90% confidence levels, and calculating upper and lower limits of simulated water quality data.

[0037] Starting from the inconsistency background, the present invention considers multiple inconsistency inputs through scenario simulation, and then obtains point predictions of water quality data representing inconsistency; secondly, by considering the uncertainty in the representativeness of water quality samples, interval predictions are made for water quality data. The water quality simulation data obtained by the present invention comprehensively considers the impact of inconsistency and the uncertainty in sample representativeness, and has certain practical and theoretical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is a step diagram of a method for predicting water quality index intervals in a non-consistent scenario provided by an embodiment of the present invention;

[0039] Figure 2 is a schematic diagram of the inconsistency comprehensive diagnosis result of the measured runoff sequence provided by an embodiment of the present invention;

[0040] Figure 3 is a correlation heat map between runoff and water quality series provided according to an embodiment of the present invention;

[0041] Figure 4is a schematic diagram of a runoff simulation sequence under various inconsistent scenarios provided by an embodiment of the present invention;

[0042] Figure 5 is a schematic diagram of water quality simulation results under scenario 1 provided in an embodiment of the present invention;

[0043] Figure 6 is a schematic diagram of water quality simulation results under scenario 2 provided in an embodiment of the present invention;

[0044] Figure 7 is a schematic diagram of water quality simulation results under scenario 3 provided in an embodiment of the present invention;

[0045] Figure 8 is a schematic diagram of water quality simulation results under scenario 4 provided in an embodiment of the present invention;

[0046] Fig. 9 is a schematic diagram of water quality simulation results under scenario 5 provided in an embodiment of the present invention;

[0047] Fig.10 is a schematic diagram of water quality simulation results under scenario 6 provided in an embodiment of the present invention;

[0048] Fig.11 It is a schematic diagram of point and interval prediction of four water qualities at a 95% confidence level in scenario 1 provided according to an embodiment of the present invention;

[0049] Fig.12 It is a schematic diagram of point and interval prediction of four water quality items at a 90% confidence level in scenario 1 provided according to an embodiment of the present invention. DETAILED DESCRIPTION

[0050] The present invention starts from the inconsistency background, simulates a variety of inconsistency inputs through scenarios, and then obtains point predictions of water quality data representing inconsistency; secondly, by analyzing the uncertainty in water quality samples, interval predictions are made for water quality data. In specific implementation, firstly, the inconsistency background of runoff data is clarified through inconsistency diagnosis, and 8 different probability distribution models are used at the same time. The water quality indicators that need to be simulated are determined according to actual needs, and the best probability distribution suitable for the water quality data and runoff data in the study area is determined through the goodness of fit test; secondly, based on the autocorrelation of runoff, the runoff is randomly simulated through the AR model, and then 6 different runoff inconsistency scenarios are generated using inconsistency scenario simulation technology; finally, a two-dimensional model is constructed through the Copula function, and different inconsistency runoff scenarios are used as input to simulate the water quality changes under different inconsistency scenarios, and the residuals of the simulated data and the observed data in the actual measurement period are sampled with replacement using the Bootstrap method, and finally the point prediction and interval prediction of the water quality simulation data under different confidence levels are obtained.

[0051] The specific implementation of the present invention is described in detail below with reference to the accompanying drawings.

[0052] Figure 1 The present invention provides a method for predicting water quality index intervals in an inconsistent scenario, comprising the following steps:

[0053] Step S100: Obtain the runoff time series and water quality time series during the measurement period and perform comprehensive diagnosis of inconsistency;

[0054] In the present invention, it is first necessary to clarify the inconsistency background of runoff data to provide a theoretical basis for subsequent steps. The present invention adopts a comprehensive test method to perform a stationarity test on the runoff time series and the water quality time series to achieve a comprehensive diagnosis of inconsistency, which mainly includes:

[0055] 1) Obtain the changing trend of each time series through the MK trend analysis method, including: using the standard normal distribution for hypothesis testing, and judging the trend change of the time series by the critical value of the MK statistic being greater than or less than the 95% confidence level;

[0056] The algorithms for MK trend analysis statistics include:

[0057] ,in, represents the intermediate statistic of the MK trend test, and Indicates the sequence number of the time series, is the ordinal length, , Assign serial numbers to the runoff or water quality values;

[0058] in, , The data relationship is expressed as:

[0059] ;

[0060] Furthermore, ,

[0061] , express The variance of Represents the result statistic of the MK trend test.

[0062] 2) Determine the inconsistency background of each time series by using the MK mutation test method, including: taking a 95% confidence level and determining the forward statistic of the test quantity curve and the backward statistics Whether the correlations intersect within the confidence interval. If so, it is judged that the time series has a significant mutation at the intersection, showing inconsistency;

[0063] The specific calculation is expressed as:

[0064] ,

[0065] ,

[0066] ,

[0067] ,

[0068] ,

[0069] ,

[0070] in, represents the intermediate statistic of the MK mutation test, and represent the mean and variance of the series respectively, and Represent the forward statistic and backward statistic of the MK mutation test, respectively.

[0071] 3) The correlation degree between the runoff sequence and the water quality sequence is clarified through correlation analysis. The present invention adopts the correlation coefficient method to perform correlation analysis, including: when the correlation coefficient is greater than 0, it indicates that there is a positive correlation between the two sequences, otherwise there is a negative correlation; the closer its absolute value is to 1, the greater the degree of correlation.

[0072] This patent uses three different correlation coefficients, Pearson, Kendall and Spearman, to perform correlation analysis on runoff series and various water quality series.

[0073] The Pearson analysis method is expressed as: ,

[0074] Kendall's method is expressed as: ,

[0075] The Spearman method is expressed as: ,

[0076] In the formula, , and They represent Pearson correlation coefficient, Kendall correlation coefficient, and Spearman correlation coefficient respectively; and Respectively represent the length time series, and Represent their means respectively; C and D represent the number of element logs with and without consistency in the two time series respectively; and Respectively represent The order of the values ​​in the sequence. and Respectively represent the mean of the sequence numbers.

[0077] The present invention provides a specific case: monthly runoff data from a local hydrological station from October 2007 to September 2018 and four monthly water quality data during the same period were used: potassium permanganate index (CODMn, mg / L), dissolved oxygen (DO, mg / L), ammonia nitrogen (NH4N, mg / L), and pH value (PH).

[0078] First, the case data is tested for inconsistency. The results are as follows: Figure 2 As shown, it shows that the runoff has inconsistent changes in many places during the measurement period, that is, the inconsistent background actually exists.

[0079] During the inconsistency test, the runoff data and the four water quality data were subjected to correlation analysis. The results are as follows: Figure 3 As shown, the runoff sequence is negatively correlated with the CODMn sequence, NH4N sequence, and PH sequence, and positively correlated with the DO sequence.

[0080] Step S110: using a variety of probability distribution models to fit the runoff time series and the water quality time series respectively, and determining the best probability distribution model according to a goodness of fit test;

[0081] In this step, various probability distribution characteristics of runoff time series and water quality time series are defined, including eight different probability distributions: extreme value distribution (EV), exponential distribution (EXP), gamma distribution (GAM), generalized extreme value distribution (GEV), generalized Pareto distribution (GP), lognormal distribution (LOGN), normal distribution (NORM), Weibull distribution (WBL), etc.

[0082] Subsequently, the optimal probability distribution model for each time series was determined based on three different goodness-of-fit tests, including the Kolmogorov-Smirnov (KS) test, the root mean square error criterion (RMSE), and the Akaikeinformation criterion (AIC information criterion).

[0083] Specifically, the EV distribution satisfies the probability density function: ;

[0084] EXP distribution satisfies the probability density function: ;

[0085] The GAM distribution satisfies the probability density function: ;

[0086] The GEV distribution satisfies the probability density function: ;

[0087] GP distribution satisfies the probability density function: ;

[0088] The LOGN distribution satisfies the probability density function: ;

[0089] NORM distribution satisfies the probability density function: ;

[0090] The WBL distribution satisfies the probability density function: .

[0091] The basis for determining the best probability distribution model based on the goodness of fit test includes: when the statistic of the KS test is equal to 1, it means that the distribution characteristics meet the hypothesis; the smaller the RMSE value, the better the fit; the smaller the AIC value, the better the fit.

[0092] Specific:

[0093] The KS test statistic is , when it is equal to 1, it means that the assumed distribution characteristics are met at a given significance level, otherwise it is not met, which is expressed as: , where represents the cumulative empirical distribution function of the sequence, represents the empirical frequency;

[0094] The smaller the RMSE value, the better the fit, which is expressed as: , where represents the empirical frequency, represents the theoretical frequency;

[0095] The smaller the AIC value, the better the fit, which is expressed as: , where Indicates the number of model parameters.

[0096] According to the case provided by the present invention, three different goodness of fit test methods, such as AIC information criterion, RMSE, and KS test, are used to determine the optimal probability distribution model of the runoff sequence and the four water quality sequences. The distribution results are shown in Table 1:

[0097] Table 1

[0098] The results show that the optimal probability distribution of runoff sequence is GP, the optimal probability distribution of CODMn sequence is LOGN, the optimal probability distribution of DO sequence is GEV, the optimal probability distribution of NH4N sequence is LOGN, and the optimal probability distribution of PH sequence is EV.

[0099] Step S120: constructing an AR model, realizing random simulation of runoff, and outputting a simulated runoff sequence; adjusting the simulated runoff sequence to obtain an inconsistent scenario;

[0100] Since runoff has autocorrelation and randomness, the AR model is used to perform random simulation of runoff.

[0101] The AR model consists of a random disturbance term The weighted sum of some items of the time series is composed of:

[0102] ,in, is the simulated runoff value at time t, is the autoregressive order, is the autoregression coefficient, is the noise term;

[0103] The runoff simulation values ​​at each moment constitute the runoff simulation sequence.

[0104] After obtaining the runoff simulation sequence, the simulated runoff sequence is adjusted to generate a variety of inconsistent scenarios.

[0105] In the present invention, adjusting the simulated runoff sequence refers to: simulating non-consistent scenarios by means of statistical factor adjustment method and seasonal variation of runoff; non-consistent scenarios include: mean increase (scenario 1), mean decrease (scenario 2), Rising (Scene 3), Decline (scenario 4), season advance (scenario 5), season lag (scenario 6);

[0106] Among them, the statistical factor adjustment method is used to simulate the inconsistency scenario according to the impact of the inconsistency condition on the statistical characteristic parameters of the sequence, and the scenario includes mean inconsistency and Inconsistency;

[0107] When generating the mean scenario, the simulation sequence obtained by the AR model can be enlarged or reduced at the 5% level while keeping other parameters unchanged, so as to obtain the runoff input under the mean inconsistency scenario: ,in, and They are the runoff sequence under the mean inconsistency scenario and the simulated runoff sequence obtained by the AR model, is the adjustment factor;

[0108] generate In the scenario of keeping the mean unchanged, the simulated sequence obtained by the AR model is The value is enlarged or reduced at the 5% level, so that The runoff input under the non-consistent scenario is expressed as: ,in, express Runoff sequences under non-consistent scenarios; is the mean runoff value.

[0109] Seasonal variation of runoff mainly focuses on the seasonal changes of runoff time series, which is used to reflect the impact of consistency conditions on runoff seasonality. The present invention mainly generates scenarios of forward inconsistency and lagging inconsistency.

[0110] Specifically, seasonal variation of runoff is expressed by the cumulative time distribution function ( ) is adjusted on the time scale, and the reconstruction model of runoff under seasonal inconsistency conditions is realized by adjusting the seasonal factors. The reconstruction model is expressed as:

[0111] ,

[0112] ,

[0113] ,

[0114] in, For the time series The cumulative time distribution function, is the length of the time series, and They represent the measured runoff time series and the runoff time series under the seasonal inconsistency scenario, respectively. , and They represent the cumulative time distribution functions of the triangular inflow process lines under the inconsistency scenario, before the inconsistency change, and after the inconsistency change, respectively.

[0115] For the case provided by the present invention, the maximum autoregressive order is determined by using AIC and Bayesian information criterion, and then the runoff AR model is constructed. The runoff sequence is simulated by the AR model, and the length of the simulation sequence is set to 100. After obtaining the runoff simulation sequence, the simulated runoff sequence is adjusted by the statistical factor adjustment method and the seasonal variation method, and then 6 runoff inconsistency scenarios are obtained (such as Figure 4 Among them, scenarios 1-4 represent the statistical characteristics of runoff series (mean, ) inconsistency, with a variation of 5%; Scenarios 5 and 6 represent seasonal inconsistency of runoff, representing advance and lag respectively.

[0116] After the six inconsistent scenarios are constructed, the simulation sequence of water quality can be generated in the two-dimensional joint Copula function model.

[0117] Step S130: constructing a two-dimensional joint Copula function model, inputting the inconsistency scenario into the two-dimensional joint Copula function model, and outputting a simulation sequence set for water quality prediction;

[0118] The Copula function can study the marginal distribution of multiple variables and the correlation structure between them separately, and has no restrictions on the type of marginal distribution. It can describe both nonlinear and asymmetric correlations. The Copula function can provide a flexible and effective simulation method for water quality prediction based on the complex relationship between variables.

[0119] The Copula functions used in the present invention include three types: Gumbel Copula, Clayton Copula and Frank Copula:

[0120] The Gumbel Copula is expressed as: ,

[0121] The Clayton Copula is expressed as: ,

[0122] Frank Copula expresses it as: ,in, It is the parameter of the Copula function.

[0123] Based on Sklar's theorem, let For marginal distribution and The joint distribution function of Satisfying formula (34), if and Continuous, then It is uniquely determined.

[0124] Therefore, in the present invention, the water quality index sequence used in the simulation is first determined. ;run-off ;

[0125] Determine the marginal distribution of the runoff and water quality index series, expressed as: ;

[0126] Select an appropriate Copula function to describe the correlation structure between runoff and water quality indicators.

[0127] Finally, based on the optimal Copula function and the constructed two-dimensional joint Copula function model, the inconsistent scenarios were input into the model to obtain the simulation sequence of various water qualities.

[0128] In the case provided by the present invention, a non-consistent two-dimensional water quality joint random simulation model is constructed using the Copula function. Three different Copula functions are used to fit the runoff data and water quality data respectively, and FrankCopula is determined to be the optimal Copula function.

[0129] Furthermore, by inputting the six inconsistent scenarios into the constructed Frank Copula model, the simulation results of the four water quality data from scenario 1 to scenario 6 can be obtained, as shown in Figures 5 to 10 shown.

[0130] Step S140: Obtain the measured runoff during the measurement period, input the measured runoff into the two-dimensional joint Copula function model, obtain a simulation sequence set of water quality, extract the residual set between the simulated values ​​and the measured values ​​of water quality points in the measurement period in the simulation sequence set, extract the residual sample set, and realize the interval prediction of water quality indicators based on the residual sample set.

[0131] Specifically, firstly, the point simulation values ​​of each water quality data in the measured period are obtained through Copula function simulation, and the residual sample set with the measured value is calculated;

[0132] Next, the Bootstrap method was used to extract the residual set between the simulated values ​​and the measured values ​​of the water quality points during the measurement period with replacement. times, get Specifically, the Bootstrap method does not need to make assumptions about the overall distribution. It only resamples based on the original sample sequence and continuously predicts the parameter values ​​of the extracted samples, thereby inferring the parameter characteristics of the unknown population and quantitatively describing the uncertainty of parameter prediction. ,set up , and its overall distribution parameter is , the Bootstrap method can be expressed as:

[0133] ,

[0134] ,

[0135] In the formula, Represents the Bootstrap statistic between the predicted population parameter and the difference between the population distribution parameter; and Indicates Parameter predictions for reproduction samples and populations; Indicates that for each The resulting parameter predictions are approximate.

[0136] Finally, the interval prediction of water quality index is realized based on the residual sample set: that is, the new sample set is sampled at the confidence level of 95% and 90% respectively to obtain the interval prediction of each water quality simulation series. A new residual sample set is sampled again at the confidence levels of 95% and 90%, respectively, to obtain the upper and lower confidence limits of the simulated values ​​of each water quality series at the confidence levels of 95% and 90%, and finally the interval prediction of each water quality simulation value can be obtained.

[0137] In the case of the present invention, firstly, a simulation sequence of four water quality data in the measured period was made by constructing a two-dimensional Copula water quality simulation model, and then a residual set between the simulated value and the measured value of the water quality point was obtained. Then, the residual set was extracted 1000 times with replacement by the Bootstrap method, and 1000 sets of new residual sample sets were obtained. Taking 95% and 90% confidence levels, 1000 sets of regenerated residual samples were sampled, and the upper and lower bounds of the residuals of the water quality simulation values ​​under the corresponding confidence levels were obtained, and then the interval prediction of the water quality simulation was obtained. The prediction results are shown in Table 2:

[0138] Table 2

[0139] Confidence Level CODMn (mg / L) DO (mg / L) NH4N (mg / L) PH 95% upper bound 0.13 0.40 0.23 0.10 95% lower bound -0.38 -0.14 -0.26 -0.26 90% upper bound 0.09 0.38 0.20 0.08 90% lower bound -0.33 -0.09 -0.22 -0.23

[0140] Among them, at the 95% and 90% confidence levels, the point prediction and interval prediction results of the four water quality data simulations in scenario 1 are as follows: Fig.11 and Fig.12 shown.

[0141] With respect to the currently commonly used technical solutions for water quality prediction, the present invention fully considers the impact of non-consistent conditions on water quality simulation and provides quantitative simulation results in specific scenarios. In the simulation environment, inconsistent scenario simulation was adopted, and 6 different representative inconsistent scenarios were obtained by using statistical factor adjustment and seasonal variation, and these inconsistent scenarios were used as input sequences for water quality simulation, overcoming the basic deficiency of previous studies that did not consider scenario simulation and directly carried out subsequent water quality simulation; further, the present invention takes inconsistent conditions as the main consideration for water quality simulation, takes 6 kinds of inconsistent scenario data as input, and obtains the water quality changes under inconsistent conditions through a two-dimensional water quality simulation model, and compares them with the simulation sequence when no inconsistency occurs, improving the method of directly using historical data or data consistent with historical data as input for simulation in previous studies, and overcoming the defect that the results of previous schemes cannot represent the actual situation; on the other hand, the present invention fully considers the uncertainty in the representativeness of water quality samples, and quantifies it in the form of point prediction and interval prediction, resamples the actual residual sample set through the Bootstrap method, and superimposes the obtained regenerated residual samples on the water quality point prediction, and obtains the interval prediction of water quality, which improves the accuracy of the prediction results compared with the water quality prediction in the traditional technical scheme that does not consider inconsistent scenarios.

[0142] The above disclosures are only several specific embodiments of the present invention; however, the present invention is not limited thereto, and any changes that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A water quality index interval prediction method under non-consistent scenarios, characterized in that: The following steps are involved: Obtain the runoff time series and water quality time series during the measured period and perform comprehensive judgment on inconsistency; Using a variety of probability distribution models, respectively fitting the runoff time series and the water quality time series, and determining the best probability distribution model based on a goodness of fit test; Construct an AR model to realize random simulation of runoff and output a simulated runoff sequence; adjust the simulated runoff sequence to obtain an inconsistent scenario; the AR model consists of a random disturbance term ε t The weighted sum of some items of the time series is composed of: Among them, x t is the simulated runoff value at time t, p is the autoregressive order, a j is the autoregression coefficient, ε t is the noise term; the adjustment of the simulated runoff sequence refers to: simulating inconsistent scenarios through statistical factor adjustment method and seasonal variation of runoff; the inconsistent scenarios include: mean increase, mean decrease, C v Rising, C v Decline, season advance, season lag; Constructing a two-dimensional joint Copula function model, inputting the inconsistency scenario into the two-dimensional joint Copula function model, and outputting a simulation sequence set for water quality prediction; Obtain the measured runoff during the measurement period, input the measured runoff into the two-dimensional joint Copula function model, obtain a simulated sequence set of water quality, extract a residual set between the simulated value and the measured value of the water quality point during the measurement period to extract a residual sample set, and realize the interval prediction of water quality indicators according to the residual sample set; The statistical factor adjustment method is used to simulate inconsistent scenarios according to the impact of inconsistent conditions on the statistical characteristic parameters of the sequence, and the scenarios include mean inconsistency and C v Inconsistency; when generating mean scenarios, in, and are the runoff sequence under the mean inconsistency scenario and the simulated runoff sequence obtained by the AR model, K is the adjustment factor; C v In the scenario of , while ensuring that the mean remains unchanged, in, Represents C v Runoff sequences under non-consistent scenarios; is the mean runoff value; The seasonal variation of runoff is used to reflect the impact of inconsistent conditions on the seasonality of runoff, and the reconstruction of runoff under seasonal inconsistent conditions is achieved through the cumulative time distribution function.

2. The water quality index interval prediction method according to claim 1, characterized in that: The execution of the inconsistency comprehensive judgment includes: The MK trend analysis method is used to obtain the changing trend of each time series, including: using the standard normal distribution to perform hypothesis testing and judging the trend change of the time series by the critical value of the MK statistic being greater than or less than the 95% confidence level; The inconsistency background of each time series is determined by the MK mutation test method, including: taking a 95% confidence level, determining whether the test quantity curve and the confidence interval intersect. If they intersect, it is determined that the time series has a significant mutation at the intersection, showing inconsistency; The correlation degree between the runoff sequence and the water quality sequence is obtained through correlation analysis, including: when the correlation coefficient is greater than 0, it means that there is a positive correlation between the two sequences, otherwise there is a negative correlation; the closer its absolute value is to 1, the greater the degree of correlation.

3. The water quality index interval prediction method according to claim 1, characterized in that: Using a variety of probability distribution models includes: determining the probability distribution characteristics of the runoff time series and each water quality time series, the probability distribution characteristics including: extreme value distribution, exponential distribution, gamma distribution, generalized extreme value distribution, generalized Pareto distribution, lognormal distribution, normal distribution and Weibull distribution; The goodness of fit includes: AIC value, RMSE value and KS test statistics; The basis for determining the best probability distribution model according to the goodness of fit test includes: when the statistic of the KS test is equal to 1, it means that the distribution characteristics of the hypothesis are met; the smaller the RMSE value, the better the degree of fit; the smaller the AIC value, the better the degree of fit.

4. The water quality index interval prediction method according to claim 1, characterized in that: The reconstruction model of runoff seasonal variation under seasonal inconsistency conditions is expressed as: TDF'(i)=TDF(i)-TDF tri1 (i)+TDF tri2 (i),i=1,2,…,n, Where TDF(j) is the jth cumulative time distribution function in the time series, n is the length of the time series, Q i and Q' i represent the measured runoff time series and the runoff time series under the seasonal inconsistency scenario, TDF'(i), TDF tri1 (i) and TDF tri2 (i) represents the cumulative time distribution function of the triangular inflow process line under the inconsistency scenario, before the inconsistency change, and after the inconsistency change.

5. The water quality index interval prediction method according to claim 1, characterized in that: The constructing of the two-dimensional joint Copula function model comprises: Determine the water quality index sequence v used in the simulation; runoff u; Determine marginal distributions of runoff and water quality indicator series; Select the Copula function to describe the correlation structure between runoff and water quality indicators.

6. The water quality index interval prediction method according to claim 1, characterized in that: The realization of interval prediction of water quality indexes according to the residual sample set refers to: performing interval prediction of simulated residuals at 95% and 90% confidence levels, and calculating upper and lower limits of simulated water quality data.

7. The water quality index interval prediction method according to claim 1, characterized in that: When extracting the residual sample set, the Bootstrap method is used to extract the residual set between the simulated value and the measured value of the water quality point in the measured period N times with replacement to obtain N groups of new residual sample sets.

Images