A parameter optimization method for nonlinear Muskingum model based on robust differential theory
Through the method based on the anti-difference differential theory, it is converted into linear parameter solution problem, and the problem of initial parameter dependence and slow solution speed in the nonlinear Mastin root model parameter optimization is solved, and the rapid and stable optimization and robustness of model parameters are achieved. .
Patent Information
- Application Number
- CN202410154126.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-02-02
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-02-02
AI Technical Summary
The existing nonlinear Mastin root model parameter optimization method relies on the initial parameter value selection, which is easy to fall into local superiority, slow solution speed, and fail to effectively resist the influence of rough errors in the measured data, resulting in a decrease in the reliability of the parameter rate determination result. .
Using a method based on the differential resistance theory, the nonlinear Mastin root model is constructed, and the sum of squares of the error is set as the objective function. The differential resistance theory is used to convert it into linear parameters to solve the problem, reduce the dependence on the initial parameters, and minimize the The two-multiple principle solves model parameters and further improves the parameter optimization direction to resist data errors.
The rapid and stable optimization of model parameters is achieved, local superiority traps are avoided, the robustness of parameter rate determination is improved, and the optimization effect is ensured in the presence of rough errors.
Smart Images

Figure CN118070502B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of model parameter optimization, and in particular relates to a nonlinear Muskingum model parameter optimization method based on robust differential theory. Background Art
[0002] River flood routing can be done by hydraulics and hydrology. The hydraulic method is mainly based on the accurate solution of the Saint-Venant equations, and the application effect is better in river sections with relatively accurate river topography and riverbed observation data. However, most rivers do not have cross-sectional observation data, and the hydraulic method is difficult to apply. The Muskingum model in hydrology has become one of the most widely used tools in the evolution of natural river channels and river floods due to its simplicity. The Muskingum model is divided into linear Muskingum and nonlinear Muskingum. The channel storage equation in the linear Muskingum model is linear, which is impractical in some cases, especially in the middle and lower reaches of large and medium-sized riverbeds, where the river slope is small and the water flow velocity increases rapidly with the increase of water level, that is, there is a nonlinear effect. In order to better simulate the change process of river flow, it is necessary to use a nonlinear Muskingum model. The key issue in the application of the nonlinear Muskingum model is the determination of the model parameters k (channel storage coefficient), x (flow weight factor) and m (channel storage equation nonlinear index). Due to the high nonlinearity of the tank storage equation, the optimization of model parameters becomes more complicated and difficult.
[0003] At present, many scholars have studied the parameter optimization of the nonlinear Muskingum model. Some scholars proposed to use the trial and error method for parameter estimation, but it takes a long time; other scholars used the nonlinear programming method to obtain better results than the trial and error method, but when solving it, it must assume appropriate initial values to ensure that the results can reach the global optimum, otherwise the solution will be trapped in the local optimum. Some scholars use the Marqand algorithm to estimate the parameters of the nonlinear Muskingum model. It also needs to assume reasonable initial values to obtain satisfactory results, and it cannot guarantee the global optimum. Some scholars use genetic algorithms (GA) to optimize the parameters of the nonlinear Muskingum model. The estimation effect of genetic algorithms is better than the above traditional optimization methods, and it does not require the initial parameter guess values to be close to the optimal value. However, the parameters such as the crossover rate and mutation rate involved in the genetic algorithm need special attention, and the genetic algorithm converges slowly. Some scholars also apply the harmony search (HS) to the optimization of the parameters of the nonlinear Muskingum model. This method minimizes the sum of squares of the error between the calculated flow and the observed flow, obtains the optimal parameter value, and does not require the initial parameter guess to be close to the optimal. However, like the GA algorithm, some parameters of the HS algorithm itself need to be given carefully. Other scholars have used the BFGS technique to calibrate the parameters of the nonlinear Muskingum model and found the best parameter values compared with previous results in terms of the sum of squared errors between the calculated and observed flow rates. However, this requires careful assumptions and complex calculations on the initial parameter values.
[0004] In summary, the existing parameter optimization methods for the nonlinear Muskingum model have the following problems: (1) The initial parameter values of the nonlinear Muskingum model or some parameters of the solution algorithm itself need to be given accurately, otherwise the global optimal value cannot be found; (2) The optimization rate is slow, which is prone to premature convergence and falling into local optimal values; (3) The existing optimization methods all rely on the information provided by the error square sum objective function to search for the optimal parameters, which belongs to the linear parameter calibration method with the error square sum as the objective function. This type of method has no problem in linear model parameter optimization, but when used for nonlinear model parameter optimization, it will increase irrelevant local optimal value solutions; (4) The existing optimization methods do not consider the quality of the data used for parameter calibration. When the actual observation data contains errors, especially gross errors, the reliability of the parameter calibration results will be reduced, and "false" optimal values will be found. Summary of the invention
[0005] The object of the present invention is to provide a nonlinear Muskingum model parameter optimization method based on robust differential theory to solve the above technical problems.
[0006] To achieve the above object, the present invention provides the following technical solutions:
[0007] The present invention discloses a nonlinear Muskingum model parameter optimization method based on robust differential theory, the method comprising the following steps:
[0008] Step 1: Construct a nonlinear Muskingum model: Collect measured hydrological data of the study area, including the measured river inflow sequence I t (t=0,1,2,……n) and the measured river section outflow sequence obQ t (t=0,1,2,……n), n is the length of the measured sample sequence, and a nonlinear Muskingum model is constructed;
[0009] The nonlinear Muskingum model consists of a water balance equation and a nonlinear tank storage equation, where the water balance equation is:
[0010]
[0011] The nonlinear tank storage equation is:
[0012]
[0013] In the formula, S t is the storage volume of the river section at time t, in m 3 ;I t is the inflow flow of the river section at time t, obtained from the measured inflow sequence of the river section, in m 3 / s;Q' t is the indicative storage flow of the river section at time t, in m3 / s; K, x, m are model parameters, where K is the tank storage coefficient, ranging from 0.001 to 1, and the unit is h; x is the flow rate weight factor, ranging from -0.5 to 0.5; m is the nonlinear index of the tank storage equation, ranging from 1 to 5, and m≠1;
[0014] According to formula (1) and formula (2), the outflow flow Q of the river section at time t is obtained: t The calculation formula is:
[0015]
[0016] Q t That is, the model calculates the outflow flow of the river section at time t;
[0017] Step 2: Set the parameter optimization objective function and parameter optimization iteration termination condition:
[0018] The objective function of parameter optimization is set to be the sum of squared errors, that is,
[0019]
[0020] Right now:
[0021]
[0022] Where, obQ t It represents the measured river section outflow at time t, obtained from the measured river section outflow sequence, in m 3 / s;Q t Indicates the model calculated river section outflow at time t, in m 3 / s; n is the length of the measured sample sequence;
[0023] The parameter optimization iteration termination conditions include the convergence of the parameter optimization objective function and the maximum number of iterations. When one of the termination conditions is met during the optimization process, the parameter optimization ends.
[0024] Step 3: Generate initial model parameters: Randomly generate a set of initial model parameters (K 0 ,x 0 ,m 0 );
[0025] Step 4: Calculate the river outflow sequence through the model: collect the measured river inflow sequence I in the study area t (t=0,1,2,…n), initial model parameters (K 0 ,x 0 ,m 0 ), input into the nonlinear Muskingum model, and obtain the model-calculated river section outflow sequence Q t0 (t=0,1,2,…n);
[0026] Step 5: Optimizing model parameters: Starting from the initial model parameters (K 0 ,x 0 ,m 0 ) is started, and the parameter optimization method based on robust differential theory is used to optimize the parameters of the nonlinear Muskingum model, and the optimized model parameters (K j+1 ,x j+1 ,m j +1 ), j is the number of parameter calibration cycles, j = 0, 1, 2, ...;
[0027] Step 6: Calculate the new river section outflow sequence: The optimized model parameters (K j+1 ,x j+1 ,m j+1 ) is substituted into the nonlinear Muskingum model to obtain a new model to calculate the river section outflow sequence Q t j+1 (t=1,2,…n);
[0028] Step 7: Determine whether the parameter optimization iteration termination condition is met: Calculate the river section outflow sequence Q using the new model t j+1 (t=1,2,…n) is substituted into the objective function for calculation to determine whether the optimized model parameters meet the parameter optimization iteration termination condition. If so, the optimization process ends. Otherwise, the new model parameters (K j+1 ,x j+1 ,m j+1 ) as the initial model parameters for the next optimization process, and return to step 4 to continue the cycle until the optimized model parameters meet the parameter optimization iteration termination condition.
[0029] Furthermore, the parameter optimization objective function in step 2 converges as follows: if the objective function value does not improve significantly after one or several iterations, the iteration is stopped, that is, when formula (6) is satisfied, the parameter optimization objective function is considered to converge;
[0030]
[0031] In the formula, SSQ j+1 SSQ j are the parameter optimization objective function values obtained for the j+1th step parameter and the jth step parameter respectively; ω is the error tolerance value;
[0032] The maximum number of iterations is limited as follows: the maximum number of iterations Z is set to 5000, and if the number of iterations exceeds this value, the calibration is stopped.
[0033] Furthermore, the specific process of optimizing the parameters of the nonlinear Muskingum model using the parameter optimization method based on robust differential theory described in step 5 is as follows:
[0034] From the perspective of mathematical differentiation, the model calculates the change in the outflow flow of the river section and the change in the parameters, and there is the following differential response relationship:
[0035]
[0036] Since formula (3) is a nonlinear function, if formula (7) is directly used to obtain partial differentials according to formula (3), the specific expression of partial differentials is still a nonlinear function. The optimization of nonlinear models or functions using the square error and objective function will add irrelevant local optimal solutions to the parameters. Therefore, the differential form is used to solve formula (7) and its partial derivatives, as shown in the following formula:
[0037]
[0038] Where: j is the number of calibration method cycles, j = 0, 1, 2, ...; Q t j =Q t ([K j ,x j ,m j ],I t ), which means that the j-th step model parameters and the river section inflow flow at time t I t Substitute the outflow flow of the river section at time t into the model; Q t j+1 =Q t ([K j+1 ,x j+1 ,m j+1 ],I t ), which means that the j+1th step model parameters and the river inflow flow at time t I t Substitute the model to obtain the river section outflow at time t; the partial derivative The calculation formula is:
[0039]
[0040] Where: ΔK, Δx, Δm are the changes in model parameters;
[0041] Assume that there are n groups of measured samples {(I1, obQ1), (I2, obQ2), … (I n , obQ n )}, and substitute it into formula (8) to obtain the following model parameter solution equations:
[0042]
[0043] Because there is bound to be an error between the model-calculated river section outflow and the measured river section outflow, the above formula uses an approximate equality sign; from formula (10), it can be seen that only the model parameters (K j+1 , x j+1 , m j+1 ) is unknown, the others are all known, and the power of the parameter is 1, so the nonlinear model parameter solution problem is transformed into a linear parameter solution problem;
[0044] Rewriting formula (10) into a vector matrix form:
[0045] obQ=Q j +H(D j+1 -D j )+β (11)
[0046] Where: obQ = [obQ1, obQ2, ..., obQ n ] T ;Q j =[Q1, Q2, ..., Q n ] T ; D j+1 =[K j+1 , x j+1 , m j+1 ] T , D j+1 is the model parameter merit to be calibrated in the next step; β = [β1, β2, …, β n ] T , indicating Q j+1 Deviation from obQ; T represents the transpose of the matrix;
[0047] According to the least squares principle, the new model parameter D j+1 The formula (12) must be satisfied:
[0048] min β T β=min[(obQ-(Q j +H(D j+1 -D j ))) T (obQ-(Q j +H(D j+1 -D j )))] (12)
[0049] According to the above formula, we can get:
[0050] D j+1 =D j +(H T H) -1 H T (obQ-Qj ) (13)
[0051] When the hydrological data used to calibrate the model does not contain gross errors, the new model parameters can be obtained by formula (13); however, since there are always errors in the measured hydrological data, the parameter optimization direction of formula (13) is further improved based on the robustness theory; the improvement principle is as follows:
[0052] The robust estimation criterion is:
[0053]
[0054] Where: ρ(·) is the extreme value function, u t is a random variable; D l =D l (d1, d2, d3) = D l (K, x, m) is the estimation vector, and l is the number of parameters to be estimated. Let Ψ be the derivative of the extreme value function, then formula (14) is expressed as:
[0055]
[0056] Right now:
[0057]
[0058] Substituting formula (11) into the above formula, formula (16) can be written in vector matrix form:
[0059]
[0060] Where:
[0061] Substituting formula (11) into formula (17) yields:
[0062]
[0063] The solution is:
[0064]
[0065] Formula (19) is the formula for optimizing the parameters of the new model in the j+1th step.
[0066] The beneficial effects of the present invention are as follows: the present invention proposes a nonlinear Muskingum model parameter optimization method based on robust differential theory, the model parameter optimization result does not depend on the selection of the initial value of the parameter, the optimal value of the model parameter can be found quickly and stably, and the influence of gross errors in measured data can be resisted, thereby ensuring the robustness of the calibration method.
[0067] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation methods. BRIEF DESCRIPTION OF THE DRAWINGS
[0068] Figure 1 is a flow chart of the method of the present invention;
[0069] Figure 2 The figure shows the results of 30 groups of different initial model parameter values and calibration parameter merits;
[0070] Figure 3 The diagram of the optimization process of parameters and objective function under the first group of initial model parameter values;
[0071] Figure 4 Comparison diagram of the difference between the model calculated flow and the measured flow obtained by different calibration methods. DETAILED DESCRIPTION
[0072] The present invention discloses a nonlinear Muskingum model parameter optimization method based on robust differential theory, such as Figure 1 As shown, the method comprises the following steps:
[0073] Step 1: Construct a nonlinear Muskingum model: Collect measured hydrological data of the study area, including the measured river inflow sequence I t (t=0,1,2,……n) and the measured river section outflow sequence obQ t (t=0,1,2,……n), n is the length of the measured sample sequence, and a nonlinear Muskingum model is constructed.
[0074] The nonlinear Muskingum model consists of a water balance equation and a nonlinear tank storage equation, where the water balance equation is:
[0075]
[0076] The nonlinear tank storage equation is:
[0077]
[0078] In the formula, S t is the storage volume of the river section at time t, in m 3 ;I t is the inflow flow of the river section at time t, obtained from the measured inflow sequence of the river section, in m 3 / s;Q' t is the indicative storage flow of the river section at time t, in m 3 / s; K, x, m are model parameters, where K is the tank storage coefficient, ranging from 0.001 to 1, and the unit is h; x is the flow rate weight factor, ranging from -0.5 to 0.5; m is the nonlinear index of the tank storage equation, ranging from 1 to 5, and m≠1;
[0079] According to formula (1) and formula (2), the outflow flow Q of the river section at time t is obtained: t The calculation formula is:
[0080]
[0081] Q t That is, the model calculates the outflow flow of the river section at time t.
[0082] Step 2: Set the parameter optimization objective function and parameter optimization iteration termination condition:
[0083] Among them, the parameter optimization objective function is the sum of square errors, that is,
[0084]
[0085] Right now:
[0086]
[0087] Where, obQ t It represents the measured river section outflow at time t, obtained from the measured river section outflow sequence, in m 3 / s;Q t Indicates the model calculated river section outflow at time t, in m 3 / s; n is the length of the measured sample sequence.
[0088] The parameter optimization iteration termination conditions include the convergence of the parameter optimization objective function and the maximum number of iterations. When one of the termination conditions is met during the optimization process, the parameter optimization ends.
[0089] 1) Convergence of parameter optimization objective function: If the objective function value does not improve significantly after one or several iterations, the iteration is stopped. That is, when formula (6) is satisfied, the parameter optimization objective function is considered to have converged.
[0090]
[0091] In the formula, SSQ j+1 SSQ j are the parameter optimization objective function values obtained for the j+1th step parameters and the jth step parameters respectively; ω is the allowable error value.
[0092] 2) Maximum number of iterations: In order to avoid the calculation program falling into an infinite loop during parameter calibration, a maximum number of iterations Z needs to be set. Generally, Z is set to 5000. If the number of iterations exceeds this value, calibration is stopped.
[0093] Step 3: Generate initial model parameters: Randomly generate a set of initial model parameters (K 0 ,x 0 ,m 0 ).
[0094] Step 4: Calculate the river outflow sequence through the model: collect the measured river inflow sequence I in the study area t (t=0,1,2,…n), initial model parameters (K 0 ,x 0 ,m 0 ), input into the nonlinear Muskingum model, and obtain the model-calculated river section outflow sequence Q t 0 (t=0,1,2,…n).
[0095] Step 5: Optimizing model parameters: Starting from the initial model parameters (K 0 ,x 0 ,m 0 ) is started, and the parameter optimization method based on robust differential theory is used to optimize the parameters of the nonlinear Muskingum model, and the optimized model parameters (K j+1 ,x j+1 ,m j +1 ), j is the number of parameter calibration cycles, j = 0, 1, 2,…
[0096] From the perspective of mathematical differentiation, the model calculates the change in the outflow flow of the river section and the change in the parameters, and there is the following differential response relationship:
[0097]
[0098] Since formula (3) is a nonlinear function, if formula (7) is directly used to obtain partial differentials according to formula (3), the specific expression of partial differentials is still a nonlinear function. The optimization of nonlinear models or functions using the square error and objective function will add irrelevant local optimal solutions to the parameters. Therefore, the differential form is used to solve formula (7) and its partial derivatives, as shown in the following formula:
[0099]
[0100] Where: j is the number of calibration method cycles, j = 0, 1, 2, ...; Q t j=Q t ([K j ,x j ,m j ],I t ), which means that the j-th step model parameters and the river section inflow flow at time t I t Substitute the outflow flow of the river section at time t into the model; Qt j+1 =Q t ([K j+1 ,x j+1 ,m j+1 ],I t ), which means that the j+1th step model parameters and the river inflow flow at time t I t Substitute the model to obtain the river section outflow at time t; the partial derivative The calculation formula is:
[0101]
[0102] Where: ΔK, Δx, Δm are small changes in model parameters.
[0103] Assume that there are n groups of measured samples {(I1, obQ1), (I2, obQ2), … (I n ,obQ n )}, and substitute it into formula (8) to obtain the following model parameter solution equations:
[0104]
[0105] Because there is bound to be an error between the model-calculated river section outflow and the measured river section outflow, the above formula uses an approximate equality sign. From formula (10), we can see that only the model parameters (K j+1 , x j+1 , m j+1 ) is unknown, the others are all known, and the power of the parameters is 1, so the nonlinear model parameter solution problem is converted into a linear parameter solution problem, which can avoid the problem of increasing the number of irrelevant local optimal solutions of nonlinear functions due to the use of the error square sum objective function for parameter optimization.
[0106] Rewriting formula (10) into a vector matrix form:
[0107] obQ=Q j +H(D j+1 -D j )+β (11)
[0108] Where: obQ = [obQ1, obQ2, ..., obQ n ] T ;Q j =[Q1, Q2, ..., Q n ] T ; D j+1 =[K j+1 , x j+1 , m j+1 ] T , Dj+1 is the model parameter merit to be calibrated in the next step; β = [β1, β2, …, β n ]T, indicating Q j+1 Deviation from obQ; T represents the transpose of the matrix.
[0109] According to the least squares principle, the new model parameter D j+1 The formula (12) must be satisfied:
[0110] min β T β=min[(obQ-(Q j +H(D j+1 -D j ))) T (obQ-(Q j +H(D j+1 -D j )))] (12)
[0111] According to the above formula, we can get:
[0112] D j+1 =D j +(H T H) -1 H T (obQ-Q j ) (13)
[0113] When the quality of the hydrological data used in the calibration model is good (i.e., the hydrological data does not contain gross errors), the new model parameters can be obtained by formula (13). However, due to differences in observation instruments, observation methods, and observation personnel, there are always errors in the measured hydrological data, and sometimes even large numbers or negative values appear. In order to better improve the applicability of the method and reduce the impact of data errors on parameter calibration, the parameter optimization direction of formula (13) is further improved based on the robustness theory. The improvement principle is as follows:
[0114] The robust estimation criterion is:
[0115]
[0116] Where: ρ(·) is the extreme value function, u t is a random variable; D l =D l (d1,d2,d3)=D l (K, x, m) is the estimation vector, l is the number of parameters to be estimated. Let Ψ be the derivative of the extreme value function, then formula (14) is expressed as:
[0117]
[0118] Right now:
[0119]
[0120] Substituting formula (11) into the above formula, formula (16) can be written in vector matrix form:
[0121]
[0122] Where:
[0123] Substituting formula (11) into formula (17) yields:
[0124]
[0125] The solution is:
[0126]
[0127] Formula (19) is the formula for optimizing the parameters of the new model in the j+1th step.
[0128] Step 6: Calculate the new river section outflow sequence: The optimized model parameters (K j+1 ,x j+1 ,m j+1 ), j = 0, ... 1, and then substituted into the nonlinear Muskingum model to obtain a new model to calculate the river section outflow sequence Q t j+1 (t=1,2,…n);
[0129] Step 7: Determine whether the parameter optimization iteration termination condition is met: Calculate the river section outflow sequence Q using the new model t j+1 (t=1,2,…n) is substituted into the objective function for calculation to determine whether the optimized model parameters meet the parameter optimization iteration termination condition. If so, the optimization process ends. Otherwise, the new model parameters (K j+1 ,x j+1 ,m j+1 ) as the initial model parameters for the next optimization process, and return to step 4 to continue the cycle until the optimized model parameters meet the parameter optimization iteration termination condition.
[0130] Embodiment 1
[0131] In order to verify the effectiveness of the method described in the present invention, this embodiment uses the above method to calibrate parameters and verifies the calibration results.
[0132] The data of this embodiment is from the book Engineering hydrology written by Wilson in 1974. The reasons for selecting the above data are: (1) the data have been proved to have an obvious nonlinear relationship between the channel storage flow and the indicated storage flow; (2) Gill (1978), Tung (1985), Yoon and Padmanabhan (1993), Mohan (1997), Chu and Chang 2009) also used these data to calibrate the parameters of the nonlinear Muskingum model. Therefore, it is convenient to compare the calibration results based on the method described in the present invention with previous studies.
[0133] In order to test the stability of the nonlinear Muskingum model parameter optimization results obtained by the method of the present invention, this embodiment uses the Random function to randomly generate 30 groups of different initial model parameter values, and performs parameter optimization respectively. The optimization results are shown in Tables 1 and Figure 2 The following conclusions can be drawn from the optimization results:
[0134] (1) The parameter optimization results are not affected by different initial parameter values. The optimization results with different initial parameter values are stable and will not fall into irrelevant local optimal solutions.
[0135] (2) The parameter optimization rate is fast. In the optimization process of 30 sets of different initial parameter values, the average number of algorithm iterations is only 18, the maximum value is 31, and the minimum value is 6.
[0136] Table 1 Calibration results using 30 groups of different initial parameter values
[0137]
[0138] Where: K 0 ,x 0 ,m 0 represents the initial model parameter value; SSQ 0 represents the initial objective function value; K, x, m represents the optimized model parameter value, and SSQ represents the optimized objective function value.
[0139] In the case of the first set of initial model parameter values, the optimization process of model parameters and objective function value (SSQ) at each step is as follows: Figure 3 As shown in the figure, the method of the present invention is easy to find the global optimal solution and has a faster convergence speed. As the number of iterations increases, the closer the parameter value searched in each step is to the optimal value, the objective function value also tends to a smaller value. All parameters can evolve in the direction of the optimal parameter value, and only 31 iterations are required to find the optimal parameter value, indicating that the parameter search direction determined by the method of the present invention is correct and effective.
[0140] The outflow sequence data of the river section obtained by substituting the parameters obtained by the method of the present invention and other methods into the model is shown in Table 2.
[0141] Table 2. River outflow sequence data calculated using different calibration methods
[0142]
[0143] Columns 1 to 3 in Table 2 are measured flow data (from Wilson 1974); columns 4 to 9 are calculated flow data obtained by substituting model parameters obtained by different parameter optimization methods into the nonlinear Muskingum model: column 4 is the LSM method (Gill 1978); column 5 is the HJ+DFP method (Tung 1985); column 6 is the NONLR method (Yoon and Padmanabhan 1993); column 7 is the GA method (Mohan 1997); column 8 is the PSO method (Chu and Chang 2009); column 9 is the method described in the present invention.
[0144] The difference between the calculated flow and the measured flow using the above six methods is shown in Figure 2. Figure 4 The comparison results of the sum of square errors (SSQ) and the sum of absolute errors (SAD) between the calculated and measured flows obtained by the above six different parameter optimization methods are shown in Table 3.
[0145]
[0146] The results show that the SSQ and SAD values of the method of the present invention are the lowest, which are (SSQ=35.57, SAD=22.85), that is, among the above 6 optimization methods, the method of the present invention has higher accuracy and is more reliable and effective.
[0147] Finally, it should be noted that the above is only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred arrangement scheme, a person skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.
Claims
1. A nonlinear Muskingum model parameter optimization method based on robust differential theory, characterized in that: The method comprises the following steps: Step 1: Construct a nonlinear Muskingum model: Collect measured hydrological data of the study area, including the measured river inflow sequence I t (t=0,1,2,……n) and the measured river section outflow sequence obQ t (t=0,1,2,……n), n is the length of the measured sample sequence, and a nonlinear Muskingum model is constructed; The nonlinear Muskingum model consists of a water balance equation and a nonlinear tank storage equation, where the water balance equation is: The nonlinear tank storage equation is: S t =KQ t '=K[xI t +(1-x)Q t ] m (2) In the formula, S t is the storage volume of the river section at time t, in m 3 ;I t is the inflow flow of the river section at time t, obtained from the measured inflow sequence of the river section, in m 3 / s;Q t ' is the river section storage flow at time t, in m 3 / s; K, x, m are model parameters, where K is the tank storage coefficient, ranging from 0.001 to 1, and the unit is h; x is the flow rate weight factor, ranging from -0.5 to 0.5; m is the nonlinear index of the tank storage equation, ranging from 1 to 5, and m≠1; According to formula (1) and formula (2), the outflow flow Q of the river section at time t is obtained: t The calculation formula is: Q t That is, the model calculates the outflow flow of the river section at time t; Step 2: Set the parameter optimization objective function and parameter optimization iteration termination condition: The objective function of parameter optimization is set to be the sum of squared errors, that is, Right now: Where, obQ t It represents the measured river section outflow at time t, obtained from the measured river section outflow sequence, in m 3 / s;Q t Indicates the model calculated river section outflow at time t, in m 3 / s; n is the length of the measured sample sequence; The parameter optimization iteration termination conditions include the convergence of the parameter optimization objective function and the maximum number of iterations. When one of the termination conditions is met during the optimization process, the parameter optimization ends. Step 3: Generate initial model parameters: Randomly generate a set of initial model parameters (K 0 ,x 0 ,m 0 ); Step 4: Calculate the river outflow sequence through the model: collect the measured river inflow sequence I in the study area t (t=0,1,2,…n), initial model parameters (K 0 ,x 0 ,m 0 ), input into the nonlinear Muskingum model, and obtain the model calculation river section outflow sequence Step 5: Optimizing model parameters: Starting from the initial model parameters (K 0 ,x 0 ,m 0 ) is started, and the parameter optimization method based on robust differential theory is used to optimize the parameters of the nonlinear Muskingum model, and the optimized model parameters (K j+1 ,x j+1 ,m j+1 ), j is the number of parameter calibration cycles, j = 0, 1, 2, ...; Step 6: Calculate the new river section outflow sequence: The optimized model parameters (K j+1 ,x j+1 ,m j+1 ) is re-substituted into the nonlinear Muskingum model to obtain a new model to calculate the outflow sequence of the river section Step 7: Determine whether the parameter optimization iteration termination condition is met: Calculate the outflow sequence of the river section using the new model Substitute into the objective function for calculation to determine whether the optimized model parameters meet the parameter optimization iteration termination conditions. If so, the optimization process ends. Otherwise, the new model parameters (K j+1 ,x j+1 ,m j+1 ) as the initial model parameters for the next optimization process, and return to step 4 to continue the cycle until the optimized model parameters meet the parameter optimization iteration termination condition; The specific process of optimizing the parameters of the nonlinear Muskingum model using the parameter optimization method based on robust differential theory described in step 5 is as follows: From the perspective of mathematical differentiation, the model calculates the change in the outflow flow of the river section and the change in the parameters, and there is the following differential response relationship: Since formula (3) is a nonlinear function, if formula (7) is directly used to obtain partial differentials according to formula (3), the specific expression of partial differentials is still a nonlinear function. The optimization of nonlinear models or functions using the square error and objective function will add irrelevant local optimal solutions to the parameters. Therefore, the differential form is used to solve formula (7) and its partial derivatives, as shown in the following formula: Where: j is the number of calibration method cycles, j = 0, 1, 2, ...; It means that the j-th step model parameters and the river section inflow flow at time t I t Substitute the outflow flow of the river section at time t into the model; It means that the j+1th step model parameters and the river inflow flow I at time t are combined t Substitute the model to obtain the river section outflow at time t; the partial derivative The calculation formula is: Where: ΔK, Δx, Δm are the changes in model parameters; Assume that there are n groups of measured samples {(I1, obQ1), (I2, obQ2), … (I n ,obQ n )}, and substitute it into formula (8) to obtain the following model parameter solution equations: Because there is bound to be an error between the model-calculated river section outflow and the measured river section outflow, the above formula uses an approximate equality sign; from formula (10), it can be seen that only the model parameters (K j+1 ,x j+1 ,m j+1 ) is unknown, the others are all known, and the power of the parameter is 1, so the nonlinear model parameter solution problem is transformed into a linear parameter solution problem; Rewriting formula (10) into a vector matrix form: obQ=Q j +H(D j+1 -D j )+β (11) Where: obQ=[obQ1,obQ2,…,obQ n ] T ;Q j =[Q1,Q2,…,Q n ] T ; D j =[K j ,x j ,m j ] T ;D j+1 =[K j+1 ,x j+1 ,m j+1 ] T , D j+1 is the model parameter merit to be calibrated in the next step; β = [β1, β2, …, β n ] T , indicating Q j+1 Deviation from obQ; T represents the transpose of the matrix; According to the least squares principle, the new model parameters are D j+1 The formula (12) must be satisfied: min(β T β)=min[(obQ-(Q j +H(D j+1 -D j ))) T (obQ-(Q j +H(D j+1 -D j )))] (12) According to the above formula, we can get: D j+1 =D j +(H T H) -1 H T (obQ-Q j ) (13) When the hydrological data used to calibrate the model does not contain gross errors, the new model parameters can be obtained by formula (13); however, since there are always errors in the measured hydrological data, the parameter optimization direction of formula (13) is further improved based on the robustness theory; the improvement principle is as follows: The robust estimation criterion is: Where: r(·) is the extreme value function, u t is a random variable; D l =D l (d1,d2,d3)=D l (K, x, m) is the estimation vector, l is the number of parameters to be estimated; let Ψ be the derivative of the extreme value function, then formula (14) is expressed as: Right now: Substituting formula (11) into the above formula, formula (16) can be written in vector matrix form: Where: Substituting formula (11) into formula (17) yields: The solution is: Formula (19) is the formula for optimizing the parameters of the new model in the j+1th step.
2. The nonlinear Muskingum model parameter optimization method based on robust differential theory according to claim 1, characterized in that: The parameter optimization objective function in step 2 converges as follows: if the objective function value does not improve significantly after one or several iterations, the iteration is stopped, that is, when formula (6) is satisfied, the parameter optimization objective function is considered to converge; In the formula, SSQ j+1 SSQ j are the parameter optimization objective function values obtained for the j+1th step parameter and the jth step parameter respectively; ω is the error tolerance value; The maximum number of iterations is limited as follows: the maximum number of iterations Z is set to 5000, and if the number of iterations exceeds this value, the calibration is stopped.
