Rainfall runoff change analysis method and system, computer and storage medium

Abstract

The invention provides a rainfall runoff change analysis method and system, a computer and a storage medium, and the method comprises the following steps: constructing a feature extraction model based on deep residual learning according to drainage basin big data; extracting drainage basin feature factors to construct a drainage basin feature factor library; and constructing a Markov logic network to excavate driving factors of rainfall runoff change for knowledge representation. A deep learning grey neural network is combined with a residual learning method, a deep residual learning-based drainage basin big data feature extraction model is constructed, accurate acquisition of drainage basin feature factors is realized, and accurate analysis of rainfall runoff change driving factors is realized by constructing a Markov logic network and combining with a drainage basin feature factor library. And a basin rainfall runoff relation change rule is disclosed, and important reference is provided for accurate simulation of a flood routing process and formulation of a fine flood control scheme.

Description

Technical Field

[0001] The present invention relates to the technical field of rainfall runoff analysis, and in particular to a rainfall runoff change analysis method, system, computer and storage medium. Background Art

[0002] The changing pattern of rainfall-runoff relationship is an important basis for the selection of hydrological models, and is also an important reference for accurate simulation of flood evolution and formulation of detailed flood prevention plans.

[0003] River basin big data contains numerous and interrelated river basin characteristics, including runoff monitoring data of similar river basins, radar rainfall data, DEM (digital elevation model) and soil and vegetation data, remote sensing image data, etc., as well as basic information such as social economy, water resources scheduling plans, previous flood control and disaster relief plans, and river basin geographical characteristics. By selecting and rejecting various characteristics and extracting important river basin characteristic factors, it can provide data and driving factor support for the analysis of changes in rainfall-runoff relationship and driving factors.

[0004] Traditional feature extraction algorithms are prone to information stacking due to the large number of basin characteristic factors and their mutual correlation. They are also difficult to accurately analyze the changes in the relationship between basin rainfall and runoff and the driving factors. There are problems such as low extraction accuracy, low efficiency, poor specificity and sensitivity. Summary of the invention

[0005] In view of the deficiencies in the prior art, the purpose of the present invention is to provide a rainfall runoff change analysis method, system, computer and storage medium, aiming to solve the technical problem of low accuracy in extracting watershed characteristic factors in the prior art.

[0006] In order to achieve the above object, in a first aspect, the present invention provides a rainfall runoff change analysis method, comprising the following steps: According to the big data of the watershed, a feature extraction model based on deep residual learning is constructed, and the deep residual learning differential equation and time response formula of the model are established; Calculating an error value output by the feature extraction model, so as to update and adjust the weight of the feature extraction model in real time according to the error back propagation and additional momentum method based on the error value; The extraction depth of watershed characteristic factors is determined by the model accuracy and the posterior difference ratio, and the watershed characteristic factors are extracted to construct a watershed characteristic factor library.

[0007] According to one aspect of the above technology, after the step of building a watershed characteristic factor library, the method further includes: Constructing a first-order predicate knowledge base of domain knowledge that affects rainfall runoff, and constructing a Markov logic network based on the first-order predicate knowledge base and the Markov network, so as to mine direct driving factors of rainfall runoff changes according to the Markov logic network; Based on the Markov logic network, combined with the basin big data, the basin characteristic factor library and the first-order predicate knowledge base, the indirect driving factors of rainfall runoff changes are mined through Gibbs sampling; The mined indirect driving factors are represented by first-order predicates, stored in a first-order predicate knowledge base, and the Markov logic network is updated; Repeat the above steps of mining indirect driving factors until there are no new indirect driving factors, and then perform knowledge representation on the final direct driving factors and indirect driving factors.

[0008] According to one aspect of the above technical solution, the expression of the deep residual learning differential equation is as follows: ; In the formula, Represents the model output parameter, which is the intermediate variable of feature extraction or the final watershed characteristic factor. Represents the model input parameter, which is the basin influencing factor in the basin big data. are the differential equation coefficients corresponding to each parameter, n is the number of parameters, and d is the differential sign; The expression of the time response formula is as follows: ； ； In the formula, is the cumulative prediction result of the target data sequence obtained after a deep learning. and is a natural exponential function, which is used to describe the exponential decay or growth behavior of parameters over time. is the response value of the nth parameter at time point t, is the response value of the model output parameter at the initial time point, The response values ​​corresponding to the model input parameters.

[0009] According to one aspect of the above technical solution, the steps of using error back propagation and additional momentum method to update and adjust the weights of the model in real time specifically include: The weights of the feature extraction model are updated using the error back propagation method based on the following calculation expression: ; ; In the formula, is the weight of each layer of the network, is the network learning rate, K is the number of samples, is the error of the kth sample, and are the network input data sequence and the network data output sequence respectively, is the partial derivative.

[0010] According to one aspect of the above technical solution, the step of using error back propagation and additional momentum method to update and adjust the weights of the model in real time also specifically includes: The weights are adjusted using the additional momentum method based on the following expression: ; ; ; In the formula, To iterate to The adjusted weights of the generation time, is the current iteration number of the deep residual learning network, To iterate to The network learning rate at each generation, is the maximum number of network iterations, and The deep residual learning network is iterated to the The maximum and minimum learning rates for each generation; is the momentum coefficient, is the dynamic adjustment factor, is the weight corresponding to the output parameter of the model, is the weight corresponding to the n-1th parameter in the model input parameters, and e is a constant.

[0011] According to one aspect of the above technical solution, the step of determining the depth of watershed characteristic factor extraction by model accuracy and posterior difference ratio specifically includes: The relative residual value of the feature extraction model is calculated, and the average residual value of the feature extraction model is calculated according to the relative residual value based on the following calculation formula: ; ; In the formula, is the mean residual value, is the relative residual value, is the feature vector of the basin big data after feature extraction, is the original feature vector in the watershed big data; The first variance value of the original data in the watershed big data is calculated based on the following calculation formula, and the second variance value of the residual value of the feature extraction model is calculated based on the average residual value: ; ; In the formula, is the first variance value, is the second variance value; The model accuracy of the feature extraction model is obtained based on the average residual value based on the following calculation formula, and the posterior difference ratio is calculated based on the first variance value and the second variance value: ; ; In the formula, is the model accuracy, is the posterior difference ratio.

[0012] According to one aspect of the above technical solution, the potential function corresponding to the joint probability distribution of the Markov network is: , ; In the formula, , is the set of c-dimensional driving factors, , is the state value taken by variable X, is the potential function of the g-th group, is the state value of all variables in the g-th group, Z is the normalization factor, and v is the number of groups.

[0013] In a second aspect, the present application also provides a rainfall runoff change analysis system, comprising: The model building module is used to build a feature extraction model based on deep residual learning according to the basin big data, and establish the deep residual learning differential equation and time response formula of the model; A parameter adjustment module, used for calculating the error value output by the feature extraction model, so as to update and adjust the weight of the feature extraction model in real time according to the error back propagation and additional momentum method based on the error value; The characteristic factor module is used to determine the extraction depth of watershed characteristic factors through model accuracy and posterior difference ratio, extract watershed characteristic factors, and build a watershed characteristic factor library.

[0014] According to one aspect of the above technology, the system further includes: A Markov logic network module is used to construct a first-order predicate knowledge base of domain knowledge that affects rainfall runoff, and to construct a Markov logic network based on the first-order predicate knowledge base and the Markov network, so as to mine direct driving factors of rainfall runoff changes according to the Markov logic network; A driving factor module, for mining indirect driving factors of rainfall runoff changes through Gibbs sampling based on the Markov logic network, combined with the watershed big data, the watershed characteristic factor library and the first-order predicate knowledge base; An updating module, used for performing first-order predicate representation on the mined indirect driving factors, storing them in a first-order predicate knowledge base, and updating the Markov logic network; The representation module is used to repeat the above steps of mining indirect driving factors until there are no new indirect driving factors, and to perform knowledge representation on the final direct driving factors and indirect driving factors.

[0015] According to one aspect of the above technical solution, the parameter adjustment module is specifically used for: The weights of the feature extraction model are updated using the error back propagation method based on the following calculation expression: ; ; In the formula, is the weight of each layer of the network, is the network learning rate, K is the number of samples, is the error of the kth sample, and are the network input data sequence and the network data output sequence respectively, is the partial derivative.

[0016] According to one aspect of the above technical solution, the parameter adjustment module is also used for: The weights are adjusted using the additional momentum method based on the following expression: ; ; ; In the formula, To iterate to The adjusted weights of the generation time, is the current iteration number of the deep residual learning network, To iterate to The network learning rate at each generation, is the maximum number of network iterations, and The deep residual learning network is iterated to the The maximum and minimum learning rates for each generation; is the momentum coefficient, is the dynamic adjustment factor, is the weight corresponding to the output parameter of the model, is the weight corresponding to the n-1th parameter in the model input parameters, and e is a constant.

[0017] According to one aspect of the above technical solution, the characteristic factor module is specifically used for: The relative residual value of the feature extraction model is calculated, and the average residual value of the feature extraction model is calculated according to the relative residual value based on the following calculation formula: ; ; In the formula, is the mean residual value, is the relative residual value, is the feature vector of the basin big data after feature extraction, is the original feature vector in the watershed big data; The first variance value of the original data in the watershed big data is calculated based on the following calculation formula, and the second variance value of the residual value of the feature extraction model is calculated based on the average residual value: ; ; In the formula, is the first variance value, is the second variance value; The model accuracy of the feature extraction model is obtained based on the average residual value based on the following calculation formula, and the posterior difference ratio is calculated based on the first variance value and the second variance value: ; ; In the formula, is the model accuracy, is the posterior difference ratio.

[0018] In a third aspect, an embodiment of the present application provides a computer, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, a rainfall runoff change analysis method as described in the first aspect is implemented.

[0019] In a fourth aspect, an embodiment of the present application provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements a rainfall runoff change analysis method as described in the first aspect.

[0020] Compared with the prior art, the beneficial effects of the present invention are as follows: by combining the deep learning grey neural network with the residual learning method, a basin big data feature extraction model based on deep residual learning is constructed, and the differential equation and time response formula of the model are established; the error back propagation and additional momentum method are used to update and adjust the model threshold and weight in real time; the basin characteristic factor extraction depth is determined by the model accuracy and the posterior difference ratio, the basin characteristic factor is extracted, and the basin characteristic factor is accurately obtained to construct a basin characteristic factor library, and then a first-order predicate knowledge base is constructed through the domain knowledge of experts to construct a Markov chain. The MLN network is used to mine direct driving factors; the indirect driving factors are mined by using the obtained MLN network, combined with the basin big data, the basin characteristic factor library and the first-order predicate knowledge base, and the mined driving factors are represented by first-order predicates, stored in the first-order predicate knowledge base, and the MLN network is updated. The second part is repeated until no new indirect driving factors are found. The search process is over and the final direct and indirect driving factors are represented by knowledge to achieve accurate analysis of the driving factors of rainfall-runoff changes, reveal the changing laws of the relationship between rainfall and runoff in the basin, and provide important reference for accurate simulation of flood evolution and formulation of refined flood control plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a flow chart of the rainfall runoff change analysis method in the first embodiment of the present invention; Figure 2 It is a structural block diagram of a rainfall runoff change analysis system in a second embodiment of the present invention; Figure 3 is a schematic diagram of the hardware structure of a computer in the third embodiment of the present application; The following specific implementation manner will further illustrate the present invention in conjunction with the above-mentioned drawings. DETAILED DESCRIPTION

[0022] In order to facilitate the understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. Several embodiments of the present invention are given in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.

[0023] It should be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be a central element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be a central element at the same time. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are for illustrative purposes only.

[0024] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more of the related listed items.

[0025] Embodiment 1 See also Figure 1 , which is a rainfall runoff change analysis method in the first embodiment of the present invention, comprising the following steps: Step S100, constructing a feature extraction model based on deep residual learning according to the watershed big data, and establishing the deep residual learning differential equation and time response formula of the model.

[0026] Specifically, in this embodiment, the grey neural network of deep learning is combined with the residual learning method and applied to the feature extraction of watershed big data, and a feature extraction model of watershed big data based on deep residual learning is constructed to realize the extraction of characteristic factors in watershed big data.

[0027] The expression of the above deep residual learning differential equation is as follows: ; In the formula, Represents the model output parameter, which is the intermediate variable of feature extraction or the final watershed characteristic factor. Represents the model input parameter, which is the basin influencing factor in the basin big data. are the differential equation coefficients corresponding to each parameter, n is the number of parameters, and d is the differential sign.

[0028] The above time response formula can be expressed as follows: ; ; In the formula, is the cumulative prediction result of the target data sequence obtained after a deep learning. and is a natural exponential function, which is used to describe the exponential decay or growth behavior of parameters over time. is the response value of the nth parameter at time point t, is the response value of the model output parameter at the initial time point, The response values ​​corresponding to the model input parameters.

[0029] Step S200, calculating the error value output by the feature extraction model, so as to update and adjust the weight of the feature extraction model in real time based on the error value according to error back propagation and additional momentum method.

[0030] Furthermore, the water cycle in a river basin involves multiple complex processes such as precipitation, evaporation, infiltration, runoff, human activities, and industrial and agricultural production. The water cycle satisfies the energy conservation principle within a certain period of time. When there are samples, the square error of the deep residual learning network output is (refers to the network input data sequence With the output sequence The square of the energy difference between them is: ; Assume that the deep residual learning network contains training samples, then the overall mean square error value of the deep residual learning network output is for: .

[0031] Specifically, in this embodiment, the residual correction process of the grey neural network has a slow convergence speed and poor stability, and the weight adjustment is performed using the additional momentum method, which can be used for target data feature extraction; The steps of using error back propagation and additional momentum method to update and adjust the model weights in real time include: The weights of the feature extraction model are updated using the error back propagation method based on the following calculation expression: ; In the formula, is the weight of each layer of the network, is the network learning rate, is the partial derivative.

[0032] The weights are adjusted using the additional momentum method based on the following expression: ; ; ; In the formula, To iterate to The adjusted weights of the generation time, is the current iteration number of the deep residual learning network, To iterate to The network learning rate at each generation, is the maximum number of network iterations, and The deep residual learning network is iterated to the The maximum and minimum learning rates for each generation; is the momentum coefficient, is the dynamic adjustment factor, is the weight corresponding to the output parameter of the model, is the weight corresponding to the n-1th parameter in the model input parameters, and e is a constant.

[0033] Specifically, the above grey neural network contains n-1 input parameters, 1 output parameter, and four network layers, namely LA layer, LB layer, LC layer and LD layer. ; . Then the initial weights of the deep residual learning network are: ; In the formula, , and They represent the weights between the LC layer and LD layer, the LB layer and LC layer, and the LA layer and LB layer of the deep residual learning network respectively.

[0034] Step S300, determining the watershed characteristic factor extraction depth through the model accuracy and the posterior difference ratio, extracting the watershed characteristic factors, and constructing a watershed characteristic factor library.

[0035] Specifically, feature extraction depth is an important indicator and parameter for deep learning feature extraction. In order to make the deep learning model have the simplest network structure while meeting the accuracy required by domain experts, the model accuracy and posterior difference ratio are used to determine the feature extraction depth.

[0036] The inspection process is as follows: calculate the relative residual value of the feature extraction model, and calculate the average residual value of the feature extraction model according to the relative residual value based on the following calculation formula: ; ; In the formula, is the mean residual value, is the relative residual value, is the feature vector of the basin big data after feature extraction, is the original feature vector in the watershed big data; The first variance value of the original data in the watershed big data is calculated based on the following calculation formula, and the second variance value of the residual value of the feature extraction model is calculated based on the average residual value: ; ; In the formula, is the first variance value, is the second variance value; The model accuracy of the feature extraction model is obtained based on the average residual value based on the following calculation formula, and the posterior difference ratio is calculated based on the first variance value and the second variance value: ; ; In the formula, is the model accuracy, is the posterior difference ratio.

[0037] It can be understood that the greater the model accuracy, the stronger the feature representation ability of the extracted target data and the higher the effectiveness; the smaller the posterior difference ratio, the smaller the error of the extracted target data features. In specific implementation, you can start with a shallower feature extraction depth (a few layers), gradually increase the number of extraction layers or dimensions, and evaluate the model's accuracy and posterior difference ratio every time you add a layer or increase the depth of feature extraction. If the accuracy is significantly improved at a specific depth and the posterior difference ratio is close to stable, then the depth can be considered to be relatively ideal, avoiding excessively deep feature extraction that leads to overly complex or underfitting models.

[0038] Under the condition of meeting the accuracy required by experts, the simplest network structure is obtained through the model accuracy and posterior difference ratio, the basin characteristic factors are obtained, and the basin characteristic factor library is established using the set correlation theory.

[0039] It is easy to understand that the changing law of rainfall-runoff relationship is an important basis for the selection of hydrological models, and is also an important reference for accurate simulation of flood evolution and formulation of detailed flood prevention plans. Traditional statistical relationship learning methods such as Bayesian and neural networks need to impose restrictions such as independence and identical distribution on data, and are only suitable for analyzing simpler data objects.

[0040] At present, when analyzing the changes in the relationship between rainfall and runoff in a watershed and its driving factors, the driving factors of the causal relationship between rainfall and runoff are mainly mined based on domain knowledge, without fully considering the influence of the correlation contained in spatially heterogeneous data, resulting in low accuracy in the characterization of the changing laws of the rainfall-runoff relationship.

[0041] Step S400, constructing a first-order predicate knowledge base of domain knowledge that affects rainfall runoff, and constructing a Markov logic network based on the first-order predicate knowledge base and the Markov network, so as to mine direct driving factors of rainfall runoff changes according to the Markov logic network.

[0042] Step S500, based on the Markov logic network, combined with the watershed big data, the watershed characteristic factor library and the first-order predicate knowledge base, the indirect driving factors of rainfall runoff changes are mined through Gibbs sampling.

[0043] Step S600: Perform first-order predicate representation on the mined indirect driving factors, store them in a first-order predicate knowledge base, and update the Markov logic network.

[0044] Step S700, repeat the above steps of mining indirect driving factors until there are no new indirect driving factors, and perform knowledge representation on the final direct driving factors and indirect driving factors.

[0045] In view of the big data of watershed, the statistical relationship learning method of Markov Logic Networks (MLNs) is introduced into big data analysis to analyze the changes in the relationship between rainfall and runoff in the watershed and its driving factors. Markov logic network is a statistical relationship learning method that combines relationship representation, likelihood reasoning and machine learning. It consists of two parts: first-order predicate and Markov network.

[0046] The driving factors are divided into two categories: direct driving factors and indirect driving factors, and are processed in two stages: In the first stage, a first-order predicate knowledge base is constructed through the domain knowledge of experts, and an MLN network is constructed to mine direct driving factors; in the second stage, the obtained MLN network is used to mine indirect driving factors in combination with watershed big data, watershed characteristic factor library and first-order predicate knowledge base, and the mined driving factors are represented by first-order predicates, stored in the first-order predicate knowledge base, and the MLN network is updated. The second part is repeated until no new indirect driving factors are found, the search process ends, and the final direct and indirect driving factors are represented by knowledge.

[0047] For ease of understanding, considering the complexity of changes in the rainfall-runoff relationship, and in order to clearly describe the changes in the rainfall-runoff relationship in a watershed and the driving factors, this embodiment takes the geographical attributes of the watershed (including watershed area, river bifurcation ratio, river network density, watershed length, watershed slope, watershed elevation, watershed shape, etc.) as an example for description, and the influence of other attributes on the rainfall-runoff relationship can be generalized accordingly.

[0048] Phase 1: The domain knowledge of experts in fields such as hydrology mainly includes the geographical location of the basin (determining whether the basin is a humid basin, arid basin, or semi-humid and semi-arid basin, etc.), climate conditions, rock and soil properties, geological structure, topography and vegetation distribution, etc., and related hydrological models, such as the water volume and energy conservation in the water cycle. Based on the domain knowledge, a first-order predicate knowledge base is constructed. The first-order predicate knowledge base consists of several first-order predicate rules. The first-order predicate rules consist of four types of symbols, namely constants, variables, functions, and predicates. For example, suppose and represent the watershed area and watershed slope respectively. , and are predicates, indicating that the basin area is related to the change of rainfall-runoff relationship, the basin area is not related to the change of rainfall-runoff relationship, and the basin area and basin slope have a dual driving relationship on rainfall-runoff.

[0049] Construct a Markov logic network. A Markov logic network is Dimension variables The joint distribution model of and defined in A set of potential functions on Based on the basin characteristic factor library and the first-order predicate knowledge base, a random field of basin rainfall-runoff relationship changes affected by basin geographical attributes is constructed. .in, The basin geographical attributes that drive changes in rainfall-runoff relationships, such as basin area, river bifurcation ratio, river network density, basin length, basin slope, basin elevation, basin shape, etc. Driving factors and There is a mutual correlation between them, forming a connecting edge For example, there is a correlation between the basin area and the basin length, which constitutes an undirected graph .like If any two nodes in a subgraph are connected by an edge, the subgraph is called a complete subgraph, and the largest complete subgraph is called a clique, denoted by ,in , v is The number of clusters in the non-negative real function About the group The potential function of The potential function set is recorded as .

[0050] The joint probability distribution of a Markov network is determined by the potential function values ​​of all groups in the network: , ; In the formula, , is the set of c-dimensional driving factors, , is the state value taken by variable X, is the potential function of the g-th group, is the state value of all variables in the g-th group, Z is the normalization factor, , v is the number of groups.

[0051] Phase II: Since the change of rainfall-runoff relationship in the basin involves various data such as meteorology, water conservancy, industry and agriculture, ecology, etc., more driving factors cannot be directly obtained through the domain knowledge of experts. Under the guidance of the first-order predicate knowledge base, with the help of the basin characteristic factor library, the Gibbs sampling technology is used to sample the basin big data, and the newly obtained indirect driving factors are represented as first-order predicate forms, stored in the first-order predicate knowledge base, and the MLN network is updated. Repeat the second phase until no new indirect driving factors are found, store the final direct and indirect driving factors in the influencing factor library, update the rainfall-runoff relationship changes, and the search process ends.

[0052] Regarding the optimization of the parameters of driving factors’ influence degree: From the potential function expression, we can know that represents the potential function state of the g-th group, represents the weight of the group, that is, the influence of each driving factor on the change of rainfall-runoff relationship in the basin. Parameter learning is an important research topic in the Markov logic network system. In the Markov network, Gibbs Sampling (GS) is used to sample the basin big data and basin characteristic factors.

[0053] Characterization of the changing law of rainfall-runoff relationship: The final step of mining the changing law of rainfall-runoff relationship is the knowledge representation of the changing law of rainfall-runoff relationship. Starting from the definition of Markov logic network, it is easy to get a graph structure: when the closed atoms represented by two nodes appear in the same closed rule, there is an edge between the two nodes. The background grid is an organizational form of the formal background, reflecting the implicit relationship between the formal background. This embodiment adopts a method based on the background grid to represent knowledge.

[0054] In summary, the rainfall runoff change analysis method in the above-mentioned embodiment of the present invention combines the deep learning grey neural network with the residual learning method to construct a basin big data feature extraction model based on deep residual learning, and establishes the differential equation and time response formula of the model; adopts error back propagation and additional momentum method to update and adjust the model threshold and weight in real time; determines the extraction depth of basin characteristic factors through the model accuracy and posterior difference ratio, extracts basin characteristic factors, realizes accurate acquisition of basin characteristic factors, and constructs a basin characteristic factor library, and then constructs a first-order predicate knowledge base through the domain knowledge of experts to construct Markov logic network is used to mine direct driving factors; the obtained MLN network is used to combine the basin big data, the basin characteristic factor library and the first-order predicate knowledge base to mine indirect driving factors, and the mined driving factors are represented by first-order predicates, stored in the first-order predicate knowledge base, and the MLN network is updated. The second part is repeated until no new indirect driving factors are found. The search process ends and the final direct and indirect driving factors are represented by knowledge to achieve accurate analysis of the driving factors of rainfall-runoff changes, reveal the changing laws of the relationship between rainfall and runoff in the basin, and provide important references for accurate simulation of flood evolution and formulation of refined flood control plans.

[0055] Embodiment 2 The second embodiment of the present application also provides a rainfall runoff change analysis system, which is used to implement the embodiments and preferred implementations, and will not be repeated hereafter. As used below, the terms "module", "unit", "subunit", etc. can implement a combination of software and / or hardware for a predetermined function. Although the system described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and contemplated.

[0056] like Figure 2 As shown, the system includes: a model building module 100, a parameter adjustment module 200, a characteristic factor module 300, a Markov logic network module 400, a driving factor module 500, an updating module 600 and a representation module 700.

[0057] The model building module 100 is used to build a feature extraction model based on deep residual learning according to the basin big data, and establish the deep residual learning differential equation and time response formula of the model; The parameter adjustment module 200 is used to calculate the error value output by the feature extraction model, so as to update and adjust the weight of the feature extraction model in real time based on the error value according to the error back propagation and additional momentum method; The characteristic factor module 300 is used to determine the extraction depth of the watershed characteristic factors through the model accuracy and the posterior difference ratio, extract the watershed characteristic factors, and construct a watershed characteristic factor library; The Markov logic network module 400 is used to construct a first-order predicate knowledge base of domain knowledge that affects rainfall runoff, and to construct a Markov logic network based on the first-order predicate knowledge base and the Markov network, so as to mine direct driving factors of rainfall runoff changes according to the Markov logic network; The driving factor module 500 is used to mine the indirect driving factors of rainfall runoff changes through Gibbs sampling based on the Markov logic network, combined with the watershed big data, the watershed characteristic factor library and the first-order predicate knowledge base; The updating module 600 is used to represent the mined indirect driving factors with first-order predicates, store them in a first-order predicate knowledge base, and update the Markov logic network; The representation module 700 is used to repeat the above steps of mining indirect driving factors until there are no new indirect driving factors, and to perform knowledge representation on the final direct driving factors and indirect driving factors.

[0058] Specifically, in this embodiment, the parameter adjustment module 200 is specifically used for: The weights of the feature extraction model are updated using the error back propagation method based on the following calculation expression: ; ; In the formula, is the weight of each layer of the network, is the network learning rate, K is the number of samples, is the error of the kth sample, and are the network input data sequence and the network data output sequence respectively, is the partial derivative.

[0059] Specifically, in this embodiment, the sliding friction module 300 is specifically used for: The parameter adjustment module 200 is also used for: The weights are adjusted using the additional momentum method based on the following expression: ; ; ; In the formula, To iterate to The adjusted weights of the generation time, is the current iteration number of the deep residual learning network, To iterate to The network learning rate at each generation, is the maximum number of network iterations, and The deep residual learning network is iterated to the The maximum and minimum learning rates for each generation; is the momentum coefficient, is the dynamic adjustment factor, is the weight corresponding to the output parameter of the model, is the weight corresponding to the n-1th parameter in the model input parameters, and e is a constant.

[0060] Preferably, in this embodiment, the characteristic factor module 300 is specifically used for: The relative residual value of the feature extraction model is calculated, and the average residual value of the feature extraction model is calculated according to the relative residual value based on the following calculation formula: ; ; In the formula, is the mean residual value, is the relative residual value, is the feature vector of the basin big data after feature extraction, is the original feature vector in the watershed big data; The first variance value of the original data in the watershed big data is calculated based on the following calculation formula, and the second variance value of the residual value of the feature extraction model is calculated based on the average residual value: ; ; In the formula, is the first variance value, is the second variance value; The model accuracy of the feature extraction model is obtained based on the average residual value based on the following calculation formula, and the posterior difference ratio is calculated based on the first variance value and the second variance value: ; ; In the formula, is the model accuracy, is the posterior difference ratio.

[0061] It should be noted that each module can be a functional module or a program module, and can be implemented by software or hardware. For modules implemented by hardware, each module can be located in the same processor; or each module can be located in different processors in any combination.

[0062] Embodiment 3 The third embodiment of the present application provides a computer. It can be understood that the principles mentioned in the rainfall runoff change analysis system in this embodiment correspond to the rainfall runoff change analysis method in the first embodiment of the present application. For details of the relevant principles not described, please refer to the first embodiment and will not be elaborated here.

[0063] The computer may include a processor 81 and a memory 82 storing computer program commands.

[0064] Specifically, the processor 81 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0065] Among them, the memory 82 may include a large-capacity memory for data or commands. By way of example and not limitation, the memory 82 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 82 may include a removable or non-removable (or fixed) medium. Where appropriate, the memory 82 may be inside or outside the data processing device. In a specific embodiment, the memory 82 is a non-volatile memory. In a specific embodiment, the memory 82 includes a read-only memory (ROM) and a random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically alterable ROM (EAROM) or a flash memory (FLASH), or a combination of two or more of these. Under appropriate circumstances, the RAM can be a static random access memory (SRAM) or a dynamic random access memory (DRAM), wherein the DRAM can be a fast page mode dynamic random access memory (FPMDRAM), an extended data output dynamic random access memory (EDODRAM), a synchronous dynamic random access memory (SDRAM), etc.

[0066] The memory 82 may be used to store or cache various data files that need to be processed and / or used for communication, as well as possible computer program commands executed by the processor 81 .

[0067] The processor 81 reads and executes the computer program commands stored in the memory 82 to implement any one of the rainfall-runoff change analysis methods in the above embodiments.

[0068] In some embodiments, the computer may further include a communication interface 83 and a bus 80. Figure 3 As shown, the processor 81, the memory 82, and the communication interface 83 are connected via a bus 80 and communicate with each other.

[0069] The communication interface 83 is used to implement communication between the modules, devices, units and / or equipment in the embodiment of the present application. The communication interface 83 can also implement data communication with other components such as: external devices, image / data acquisition equipment, databases, external storage, and image / data processing workstations.

[0070] The bus 80 includes hardware, software or both, and couples the components of the computer to each other. The bus 80 includes but is not limited to at least one of the following: a data bus, an address bus, a control bus, an expansion bus, and a local bus. By way of example and not limitation, bus 80 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses or a combination of two or more of these. Where appropriate, bus 80 may include one or more buses. Although embodiments of the present application describe and illustrate a particular bus, the present application contemplates any suitable bus or interconnect.

[0071] Embodiment 4 In combination with the rainfall runoff change analysis method in the above embodiments, the fourth embodiment of the present application provides a readable storage medium. The readable storage medium stores computer program commands; when the computer program commands are executed by a processor, any rainfall runoff change analysis method in the above embodiments is implemented.

[0072] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0073] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A rainfall runoff change analysis method, characterized in that: The following steps are involved: According to the big data of the watershed, a feature extraction model based on deep residual learning is constructed, and the deep residual learning differential equation and time response formula of the model are established; Calculating an error value output by the feature extraction model, so as to update and adjust the weight of the feature extraction model in real time according to the error back propagation and additional momentum method based on the error value; The extraction depth of watershed characteristic factors is determined by model accuracy and posterior difference ratio, and the watershed characteristic factors are extracted to construct a watershed characteristic factor library.

2. The rainfall runoff change analysis method according to claim 1, characterized in that: After the step of constructing a watershed characteristic factor library, the method further comprises: Constructing a first-order predicate knowledge base of domain knowledge that affects rainfall runoff, and constructing a Markov logic network based on the first-order predicate knowledge base and the Markov network, so as to mine direct driving factors of rainfall runoff changes according to the Markov logic network; Based on the Markov logic network, combined with the basin big data, the basin characteristic factor library and the first-order predicate knowledge base, the indirect driving factors of rainfall runoff changes are mined through Gibbs sampling; The mined indirect driving factors are represented by first-order predicates, stored in a first-order predicate knowledge base, and the Markov logic network is updated; Repeat the above steps of mining indirect driving factors until there are no new indirect driving factors, and then perform knowledge representation on the final direct driving factors and indirect driving factors.

3. The rainfall runoff change analysis method according to claim 1, characterized in that: The expression of the deep residual learning differential equation is as follows: ； In the formula, Represents the model output parameter, which is the intermediate variable of feature extraction or the final watershed characteristic factor. Represents the model input parameter, which is the basin influencing factor in the basin big data. are the differential equation coefficients corresponding to each parameter, n is the number of parameters, and d is the differential sign; The expression of the time response formula is as follows: ； ； In the formula, is the cumulative prediction result of the target data sequence obtained after one deep learning. and is a natural exponential function, which is used to describe the exponential decay or growth behavior of parameters over time. is the response value of the nth parameter at time point t, is the response value of the model output parameter at the initial time point, The response values ​​corresponding to the model input parameters.

4. The rainfall runoff change analysis method according to claim 3 is characterized in that: The steps of using error back propagation and additional momentum method to update and adjust the model weights in real time include: The weights of the feature extraction model are updated using the error back propagation method based on the following calculation expression: ； ； In the formula, is the weight of each layer of the network, is the network learning rate, K is the number of samples, is the error of the kth sample, and are the network input data sequence and the network data output sequence respectively, is the partial derivative.

5. The rainfall runoff change analysis method according to claim 4 is characterized in that: The steps of using error back propagation and additional momentum method to update and adjust the weights of the model in real time also specifically include: The weights are adjusted using the additional momentum method based on the following expression: ； ； ； In the formula, To iterate to The adjusted weights of the generation time, is the current iteration number of the deep residual learning network, To iterate to The network learning rate at each generation, is the maximum number of network iterations, and The deep residual learning network is iterated to the The maximum and minimum learning rates for each generation; is the momentum coefficient, is the dynamic adjustment factor, is the weight corresponding to the output parameter of the model, is the weight corresponding to the n-1th parameter in the model input parameters, and e is a constant.

6. The rainfall runoff change analysis method according to claim 3 is characterized in that: The steps of determining the depth of watershed characteristic factor extraction by model accuracy and posterior difference ratio specifically include: The relative residual value of the feature extraction model is calculated, and the average residual value of the feature extraction model is calculated according to the relative residual value based on the following calculation formula: ； ； In the formula, is the mean residual value, is the relative residual value, is the feature vector of the basin big data after feature extraction, is the original feature vector in the watershed big data; The first variance value of the original data in the watershed big data is calculated based on the following calculation formula, and the second variance value of the residual value of the feature extraction model is calculated based on the average residual value: ； ； In the formula, is the first variance value, is the second variance value; The model accuracy of the feature extraction model is obtained based on the average residual value based on the following calculation formula, and the posterior difference ratio is calculated based on the first variance value and the second variance value: ； ； In the formula, is the model accuracy, is the posterior difference ratio.

7. The rainfall runoff change analysis method according to claim 2, characterized in that: The potential function corresponding to the joint probability distribution of the Markov network is: ， ； In the formula, , is the set of c-dimensional driving factors, , is the state value taken by variable X, is the potential function of the g-th group, is the state value of all variables in the g-th group, Z is the normalization factor, and v is the number of groups.

8. A rainfall runoff change analysis system, characterized in that: include: The model building module is used to build a feature extraction model based on deep residual learning according to the basin big data, and establish the deep residual learning differential equation and time response formula of the model; A parameter adjustment module, used for calculating the error value output by the feature extraction model, so as to update and adjust the weight of the feature extraction model in real time according to the error back propagation and additional momentum method based on the error value; The characteristic factor module is used to determine the extraction depth of watershed characteristic factors through model accuracy and posterior difference ratio, extract watershed characteristic factors, and build a watershed characteristic factor library.

9. A computer comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the rainfall runoff change analysis method according to any one of claims 1 to 7 is implemented.

10. A storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the rainfall runoff change analysis method described in any one of claims 1 to 7 is implemented.

Images