Slope stability prediction method and device

Abstract

The invention discloses a slope stability prediction method and device. The method comprises the steps of obtaining training data and establishing a PPO-PSO-BP prediction model for predicting slope stability; training data are input into the PPO-PSO-BP prediction model, training optimization is carried out on particle swarm parameters according to a strategy network and a value network to obtain an optimal strategy so as to determine optimal particles, the particle swarm parameters specifically comprise an inertia weight # imgabs0 #, an individual cognitive learning factor # imgabs1 # and a social cognitive learning factor # imgabs2 #, and the particle swarm parameters specifically comprise an inertia weight # imgabs0 #, an individual cognitive learning factor # imgabs1 # and a social cognitive learning factor # imgabs2 #. The optimal particle is specifically an optimal weight parameter and an optimal offset parameter of a BP network in the PPO-PSO-BP prediction model; and inputting the slope data at the current moment into the PPO-PSO-BP prediction model after the parameters are determined to obtain the slope stability at the prediction moment, so that the problems of poor population diversity, difficulty in convergence and falling into a local extreme value are avoided, and the prediction precision and prediction stability of the slope stability can be effectively improved.

Description

Technical Field

[0001] The invention belongs to the technical field of slope disaster assessment, and in particular relates to a slope stability prediction method and device. Background Art

[0002] Slope stability is one of the major research topics in geotechnical engineering, and is involved in many fields such as mining engineering, water conservancy engineering, and construction engineering. Slope engineering is a dynamic, open, complex, and nonlinear system, and slope stability is easily affected by various factors.

[0003] When the traditional particle swarm algorithm is used to predict slope stability, the parameters in the algorithm (inertia weight and learning factor) are usually fixed values ​​set manually. In dynamic optimization problems, it is difficult to monitor the algorithm's operating status in real time and make timely adjustments based on environmental changes, which leads to problems such as premature convergence and falling into local optimality when solving complex optimization problems.

[0004] Therefore, how to improve the prediction accuracy of slope stability and establish a more adaptable network model so that the algorithm parameters can change dynamically according to the state of particles during the iteration process, avoiding the problem of poor algorithm population diversity, difficulty in convergence, and easy to fall into local extreme values ​​caused by fixed parameters, is a technical problem to be solved by technical personnel in this field. Summary of the invention

[0005] The purpose of the present invention is to solve the technical problems in the prior art that the slope stability prediction result has low accuracy and the model has poor adaptability.

[0006] To achieve the above technical objectives, on the one hand, the present invention provides a slope stability prediction method, the method comprising: Obtain training data and establish a PPO-PSO-BP prediction model for predicting slope stability; The training data is input into the PPO-PSO-BP prediction model, and the particle swarm parameters are trained and optimized according to the strategy network and value network to obtain the optimal strategy. , the optimal particles are obtained by iteration according to the optimal strategy using the test data, and the particle swarm parameters specifically include the inertia weight , individual cognitive learning factors and social cognitive learning factors , the optimal strategy is specifically that the Actor network in the PPO algorithm outputs a probability to determine the movement mode of the particle in the search space, and the optimal particle is specifically the optimal weight parameter and the optimal bias parameter of the BP network in the PPO-PSO-BP prediction model; The slope data at the current moment is input into the PPO-PSO-BP prediction model after the parameters are determined to obtain the slope stability at the prediction moment.

[0007] Furthermore, the particle swarm parameters are trained and optimized according to the strategy network and the value network to obtain the optimal strategy π, specifically including: The current state of the agent corresponding to the PPO algorithm is determined according to the current position and current fitness value of each particle in the particle swarm. Specifically, it includes the variance of the fitness value at the current moment, the particle aggregation, the current iteration progress, and the difference between the optimal fitness value of the population at the current moment and the preset accuracy. The current state of the agent is specifically determined by the following formula: ; ; ; ; ; ; In the formula, for The variance of the fitness value at each moment is expressed as express, for The variance of the fitness value at each moment, for Moment The fitness value of a particle, for time The mean fitness value of particles, for The particle aggregation degree at the moment is express, is the search space where the particle swarm is located, for Moment The position of a particle, for The position of the central particle of the particle group at time is the current iteration progress, is the current iteration number, is the preset maximum number of iterations, is the difference between the optimal fitness value of the population at the current moment and the preset accuracy, for The optimal fitness value of the population at the moment; The strategy network is used to output the probability distribution of actions according to the current state of the agent, and actions are sampled from it. After inputting into the operating environment (particle swarm parameter optimization problem), the next state of the agent is determined , and get an instant reward After completing one iteration; The action For a three-dimensional action, : ; ; ; The three-dimensional action specifically corresponds to the particle swarm parameters, namely the inertia weight , ( ω ∈[0.4,0.9]), individual cognitive learning factor and social cognitive learning factors ; The Reward Set to: ; in, They are: ; ; ; ,when ; In the formula, for The optimal fitness value of the population at the moment, for The optimal fitness value of the population at the moment, for The variance of the fitness value at each moment, for The variance of the fitness value at each moment; for The aggregation degree of particles at a given moment, for The aggregation degree of particles at a given moment; The speed and position of the particles in the particle swarm are updated according to the action. The speed and position of the particles are updated specifically by the following formula: ; ; In the formula, for Moment The speed of a particle, for Moment The speed of a particle, For the The dimension of a particle, and for A random number in the range, For the The best historical position of a particle, is the optimal position of all particles, for Moment The position of a particle, for Moment The position of a particle; The current state of the agent ,action , Instant Rewards and the next state of the agent Form a trajectory and store the trajectory in the experience buffer; When the number of iterations reaches the maximum number of iterations of the round or the minimum fitness of the particle swarm reaches the preset accuracy, the value network is updated, and the advantage function is determined according to all trajectories in the experience buffer area through the updated value network, and then the strategy network is updated through the advantage function to end the current round.

[0008] Furthermore, the parameters of the value network are specifically calculated by the following formula: To update: ; In the formula, for Parameters of the moment value network, is the update step size of the value network parameters, for The time series difference error value, is the gradient derivative, for The state value function at the moment; Updating the policy network by the advantage function is specifically to update the objective function of the policy network by the following formula : ; ; ; In the formula, is the maximum iteration step, is the current iteration step, is the truncation function, is the mean calculation function, is the ratio of the new and old strategies, is the advantage function, is the minimum value, Yes Truncation is performed to limit the magnitude of the policy update. Based on the ratio of new and old strategies With advantage function The product of reflects the advantage of the action under the current strategy and the impact of strategy changes. Yes After truncation, the advantage function Multiply to avoid training instability due to large changes in strategy. Yes and The minimum value between the two ensures that a relatively conservative and stable update amount is selected during the policy update process; is the discount factor, To control the weight of multi-step returns in the advantage estimation, for The time difference error value under the moment decision step, is the time series difference error value before the maximum iteration step, For status The value estimation error of for When for status The estimated value of the cumulative reward after a certain number of steps, for The state value function at time , which means that Start the expectation of accumulating rewards in the future, for The state value function at the moment, which represents the state from Start the expectation of accumulating rewards in the future.

[0009] The parameters of the policy network The update is performed using the following formula: ; In the formula, for The parameters of the policy network at the decision step, is the policy network parameter update step size, for The time difference error value under the moment decision step, is the gradient derivative, is a logarithmic function, For the parameter Determined policy function, expressed in state Take action probability.

[0010] On the other hand, the present invention also provides a slope stability prediction device, the device comprising: An acquisition module is used to acquire training data and establish a PPO-PSO-BP prediction model for predicting slope stability; The training module is used to input the training data into the PPO-PSO-BP prediction model, and train the particle swarm parameters according to the strategy network and value network to obtain the optimal strategy. , the optimal particles are obtained by iteration according to the optimal strategy using the test data, and the particle swarm parameters specifically include the inertia weight , individual cognitive learning factors and social cognitive learning factors , the optimal strategy is specifically that the Actor network in the PPO algorithm outputs a probability to determine the movement mode of the particle in the search space, and the optimal particle is specifically the optimal weight parameter and the optimal bias parameter of the BP network in the PPO-PSO-BP prediction model; The prediction module is used to input the slope data at the current moment into the PPO-PSO-BP prediction model after the parameters are determined to obtain the slope stability at the prediction moment.

[0011] The present invention provides a slope stability prediction method and device. Compared with the prior art, the present invention first obtains training data and establishes a PPO-PSO-BP prediction model for predicting slope stability; then the training data is input into the PPO-PSO-BP prediction model, and the particle swarm parameters are trained and optimized according to the strategy network and the value network to obtain the optimal strategy, and the optimal particles are iterated according to the optimal strategy using the test data. The particle swarm parameters specifically include inertia weight , individual cognitive learning factors and social cognitive learning factors The optimal strategy is that the PPO algorithm Actor network outputs a probability distribution to determine the movement of particles in the search space. The optimal particles are the optimal weight parameters and the optimal bias parameters of the BP network in the PPO-PSO-BP prediction model. Finally, the slope data at the current moment is input into the PPO-PSO-BP prediction model after the parameters are determined to obtain the slope stability at the prediction moment, avoiding the problems of poor population diversity, difficulty in convergence and falling into local extreme values ​​in the prior art, which can effectively improve the prediction accuracy and prediction stability of slope stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the embodiments of this specification or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0013] Figure 1 Shown is a schematic flow chart of a slope stability prediction method provided in an embodiment of this specification; Figure 2 Shown is a schematic diagram of the structure of a slope stability prediction device provided in an embodiment of this specification; Figure 3 The figure shows the comparison results between the PSO-BP algorithm in the prior art provided in the embodiments of this specification and the solution of the present application. DETAILED DESCRIPTION

[0014] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0015] like Figure 1 The flowchart of the slope stability prediction method provided by the embodiment of this specification is shown. Although this specification provides the method operation steps or device structures shown in the following embodiments or drawings, the method or device may include more or fewer operation steps or module units after partial merger based on routine or no creative labor. In the steps or structures where there is no necessary causal relationship logically, the execution order of these steps or the module structure of the device is not limited to the execution order or module structure shown in the embodiments or drawings of this specification. When the method or module structure is applied in actual devices, servers or terminal products, it can be executed sequentially or in parallel according to the method or module structure shown in the embodiments or drawings (for example, in a parallel processor or multi-threaded processing environment, or even in a distributed processing, server cluster implementation environment).

[0016] The slope stability prediction method provided in the present embodiment can be applied to terminal devices such as clients and servers. Figure 1 As shown, the method specifically comprises the following steps: Step S101, obtaining training data and establishing a PPO-PSO-BP prediction model for predicting slope stability.

[0017] Specifically, the training data includes historical data of slopes, which include slope geometric elements (slope height, slope angle, etc.), rock and soil parameters (soil bulk density, cohesion, internal friction angle, elastic modulus, Poisson's ratio, etc.); natural factors (earthquakes, rainfall, etc.), and human factors (excavation, support, etc.). In general, the greater the slope height and the greater the slope angle, that is, the steeper the slope, the worse the slope stability; when the rock and soil bulk density is large, the cohesion is greater, and the internal friction angle is larger, the stability is better. Based on previous studies, relevant literature and engineering practice, bulk density, cohesion, internal friction angle, slope angle, slope height and pore water pressure are selected as the main indicators for measuring the stability state of the slope.

[0018] In order to verify the prediction performance of the PPO-PSO-BP algorithm, the present invention trains the model and predicts the data based on actual slope data. In order to establish a reliable slope safety factor training and testing database, 103 slope engineering examples were selected from relevant literature as training samples. The sample data is divided into two parts, 83 groups of data are used as training samples to train and learn the network, and 20 groups of data are used as test samples to verify the feasibility of the network model constructed by the present invention. The training data and test data are as follows, where bulk density, cohesion, internal friction angle, slope angle, slope angle, slope height and pore water pressure are the inputs of the model, and the safety factor of the slope is the output of the model: Table 1 Model training data ; Table 2 Model test data ;

[0019] After obtaining the training data, it is also necessary to normalize the training data using the following formula: ; In the formula, is the training data, is the normalized training data; is the maximum value of the variation range of sample data of the same factor, and the sample data of the same factor is each data, including bulk density, cohesion, internal friction angle, slope angle, slope height, pore water pressure and safety factor; The minimum value of the variation range of sample data of the same factors also includes bulk density, cohesion, internal friction angle, slope angle, slope height, pore water pressure and safety factor; .

[0020] After obtaining the normalized training data, a PPO-PSO-BP prediction model for predicting slope stability is established. BP is also known as the Back Propagation neural network. Particle Swarm Optimization (PSO) is a random global search algorithm based on swarm intelligence theory, which is established by simulating the migration and aggregation behavior of bird flocks during foraging. The algorithm has few parameters, simple calculations, and is easy to implement. PPO (Proximal Policy Optimization) is a policy gradient-based reinforcement learning algorithm used to optimize the policy network to maximize the cumulative reward. It mainly includes two networks: the policy network (used to generate policies and actions) and the value network (used to evaluate the state value).

[0021] Initialize it, set the number of hidden layers and the number of nodes in the input layer, hidden layer, and output layer.

[0022] The number of nodes in the hidden layer is determined by the following formula: ; In the formula, is the number of nodes in the network input layer. According to the input in the training data, it is bulk density, cohesion, internal friction angle, slope angle, slope height, and pore water pressure. ; m is the number of hidden layer nodes, according to the number of training data types and test results, ; n is the number of nodes in the output layer. Since the output is the safety factor of the slope, ; is the adjustment constant, take 10.

[0023] Set the activation function of the neural network and select the Sigmoid function as the activation function, as follows: ; In the formula, is the Sigmoid function, is an exponential function.

[0024] Specifically, after initializing the neural network model, a particle swarm network is established based on the parameters of the neural network model, and the population number N, the number of iterations t, and the maximum number of iterations are set. The reinforcement learning algorithm uses the Proximal Policy Optimization (PPO) algorithm to initialize the parameters of the policy network (Actor) And the parameters of the value network (Critic) , set the number of rounds The maximum number of rounds is , limited by the maximum position and maximum speed to avoid blindly searching in space, set the position range of the particle and speed range , calculate the search space : ; In the formula, are the number of nodes in the network input layer, hidden layer, and output layer, respectively.

[0025] Calculate the fitness value of each particle according to the mean square error formula : ; In the formula, For the The fitness value of a particle, Output value for the network; is the tag value.

[0026] Step S102: input the training data into the PPO-PSO-BP prediction model, and train the particle swarm parameters according to the strategy network and the value network to obtain the optimal strategy π, and use the test data to iterate according to the optimal strategy to obtain the optimal particle. The particle swarm parameters specifically include the inertia weight , individual cognitive learning factors and social cognitive learning factors The optimal strategy is specifically that the Actor network in the PPO algorithm outputs a probability to determine the movement mode of the particle in the search space, and the optimal particle is specifically the optimal weight parameter and the optimal bias parameter of the BP network in the PPO-PSO-BP prediction model.

[0027] In the embodiment of the present application, the particle swarm parameters are trained and optimized according to the strategy network and the value network to obtain the optimal strategy, specifically including: At each iteration during the training process, the current state of the agent corresponding to the PPO algorithm is determined by the current position and current fitness. , the current state of the agent Specifically includes the variance of the fitness value at the current moment , particle aggregation , Current iteration progress The difference between the optimal fitness value of the population at the current moment and the preset accuracy In the present application, the preset accuracy can be set to 0.01, or can be flexibly set by those skilled in the art according to actual conditions. The current state of the agent is specifically determined by the following formula: ; ; ; ; ; ; In the formula, for The variance of the fitness value at each moment is expressed as express, for The variance of the fitness value at each moment; for The aggregation degree of particles at the moment is expressed as express; The current iteration progress; It is the difference between the optimal fitness value of the population at the current moment and the preset accuracy. for Moment The fitness value of a particle, for time N The mean fitness value of particles, is the search space where the particle swarm is located, for Moment The position of a particle, for The position of the central particle of the particle group at time is the current iteration number, is the preset maximum number of iterations, for The optimal fitness value of the population at the moment.

[0028] The strategy network is used to output the probability distribution of actions according to the current state of the agent, and actions are sampled from it. After inputting into the operating environment (particle swarm parameter optimization problem), the next state of the agent is determined , and get an instant reward r t Then complete one iteration.

[0029] Specifically, the action For a three-dimensional action : ; ; ; The three-dimensional action specifically corresponds to the particle swarm parameter, i.e., the inertia weight , ( ω ∈[0.4,0.9]), individual cognitive learning factor and social cognitive learning factors , ( c 1, c 2∈[0,4]).

[0030] Specifically, the reward r t Set to: ; in, They are: ; ; ; ,when ; In the formula, for The optimal fitness value of the population at the moment, for The optimal fitness value of the population at the moment, for The variance of the fitness value at each moment, for The variance of the fitness value at each moment; for The aggregation degree of particles at a given moment, for t The aggregation degree of particles at a certain moment.

[0031] After obtaining the action Then the particle swarm parameters are updated, and then the particle speed and position are updated. That is to say, in the iterative process, the weight and bias of the BP network corresponding to the particle position are brought into the PPO-PSO-BP prediction model to obtain the predicted value, and then the predicted value and the label value are brought into the fitness function to obtain the fitness value of the particle. After comparing the fitness values, the optimal position of each individual particle and the global optimal position of the particle swarm are searched in space. The best position found by the i-th particle so far is recorded as , the corresponding fitness is ; At the same time, each particle shares information with other particles in the entire particle group, and the best position searched by all particles is recorded as , the corresponding fitness is : ; ; The speed and position of the particle are updated specifically by the following formula: ; ; In the formula, for Moment The speed of a particle, for Moment The speed of a particle, For the The dimension of a particle, and for A random number in the range, For the The best historical position of a particle, is the optimal position of all particles, for Moment The position of a particle, Moment The position of a particle.

[0032] Through the above operations, an iteration is completed. Then, at the end of the current iteration, the current state of the agent needs to be ,action , Instant Rewards r t and the next state of the agent A trajectory is formed and stored in the experience buffer. After the current iteration is completed, it is necessary to determine whether the number of iterations reaches the maximum number of iterations of the round or whether the minimum fitness of the particle swarm reaches the preset accuracy. When the end condition is reached, the value network is updated, and the advantage function is determined according to all the trajectories in the experience buffer through the updated value network, and then the strategy network is updated through the advantage function.

[0033] The experience buffer stores the trajectory determined by each iteration in the current training round. At the end of the current training round, both the value network and the policy network need to be updated. First, the value network is updated by minimizing the objective function loss of the value network and back-propagating the loss function with respect to the parameters. The gradient of the value network is used to update the value network, as shown in the following formula: ; In the formula, for The parameters of the value network at the moment decision step, is the update step size of the value network parameters, for The time difference error value under the moment decision step, is the gradient derivative, for The state value function at time.

[0034] The update goal of the policy network is to maximize the cumulative return. The gradient of the objective function is estimated by sampling multiple rounds of data from the experience buffer, including the state-action pairs and reward information in the buffer when calculating the advantage function and the ratio of the new and old policies. This information can help the policy network learn which actions are more likely to obtain high cumulative returns in different states. The advantage function is determined based on all the trajectories in the experience buffer through the updated value network. , the advantage function It is determined specifically by the following formula: ; ; In the formula, is the maximum iteration step, is the current iteration step, is the discount factor, To control the weight of multi-step returns in the advantage estimation, for The time difference error value under the moment decision step, is the time series difference error value before the maximum iteration step, for When for status The estimated value of the cumulative reward after a certain number of steps, for The state value function at time , which means that Start the expectation of accumulating rewards in the future, for The state value function at time.

[0035] After determining the advantage function, the objective function of the policy network is updated by the following formula : ; In the formula, is the truncation function, is the mean calculation function, is the ratio of the new and old strategies, is the advantage function, is a minimum value. This objective function can ensure that the strategies generated by two adjacent iterations will not produce large deviations.

[0036] After updating the objective function of the policy network, the parameters of the policy network are Update using the following formula: ; In the formula, for The parameters of the policy network at the decision step, is the policy network parameter update step size, for The time difference error value under the moment decision step, is the gradient derivative, is a logarithmic function, For the parameter Determined strategy.

[0037] By continuously iterating the above parameter update formula, the parameters of the policy network Will be adjusted gradually. As the number of iterations increases, the policy network parameters It will converge to a stable value. At this point, the strategy determined by the policy network is the optimal strategy .

[0038] Step S103: input the slope data at the current moment into the PPO-PSO-BP prediction model after the parameters are determined to obtain the slope stability at the prediction moment.

[0039] Specifically, the optimal strategy It enables the particles in PSO to take the optimal action sequence in the environment, and then iterate to obtain the optimal particles, corresponding to the optimal weight and bias of the BP network. The input of 20 sets of test data, including bulk density, cohesion, internal friction angle, slope angle, slope height, and pore water pressure data, is input into the PPO-PSO-BP prediction model to obtain the slope safety factor.

[0040] Comparing the conventional PSO-BP algorithm in the prior art with the solution of the present application, the parameters of the conventional PSO-BP algorithm are set as , , the results are as follows Figure 3 As shown, the prediction error of the traditional PSO-BP algorithm is 0.0839, and the prediction error of the algorithm in the present application is 0.0132, which is only 1 / 6 of the traditional PSO-BP algorithm. In addition, the correlation between the prediction value of the traditional PSO-BP algorithm and the label value is poor, which is only -0.0172. The correlation between the prediction value of the algorithm in the present application and the label value is 0.8608, which has a good correlation.

[0041] This scheme first obtains training data and establishes a PPO-PSO-BP prediction model for predicting slope stability; then the training data is input into the PPO-PSO-BP prediction model, and the particle swarm parameters are trained and optimized according to the strategy network and the value network to obtain the optimal strategy π, and the optimal particles are obtained by iteration according to the optimal strategy using the test data. The particle swarm parameters specifically include the inertia weight , individual cognitive learning factors and social cognitive learning factors The optimal strategy is that the PPO algorithm Actor network outputs a probability distribution to determine the movement of particles in the search space. The optimal particles are the optimal weight parameters and the optimal bias parameters of the BP network in the PPO-PSO-BP prediction model. Finally, the slope data at the current moment is input into the PPO-PSO-BP prediction model after the parameters are determined to obtain the slope stability at the prediction moment, avoiding the problems of poor population diversity, difficulty in convergence and falling into local extreme values ​​in the prior art, which can effectively improve the prediction accuracy and model adaptability of slope stability.

[0042] The solution provided by the present application can solve the problem that the parameter optimization process in the traditional particle swarm optimization neural network is prone to premature convergence and falling into the local optimum, resulting in low prediction accuracy and model adaptability when the slope stability is predicted. The solution of the present application can dynamically change according to the state of the particles during the training iteration process, avoiding the problem of poor algorithm population diversity, difficulty in convergence and falling into local extreme values ​​due to fixed parameters, so that the solution provided by the present application is more accurate and stable when the slope stability prediction is finally performed.

[0043] Based on the above-mentioned slope stability prediction method, one or more embodiments of this specification also provide a platform and terminal for slope stability prediction. The platform or terminal may include a device, software, module, plug-in, server, client, etc. that uses the method described in the embodiments of this specification and is combined with a device for implementing necessary hardware. Based on the same innovative concept, the system in one or more embodiments provided in the embodiments of this specification is as described in the following embodiments. Since the implementation scheme and method for solving the problem of the system are similar, the implementation of the specific system in the embodiments of this specification can refer to the implementation of the aforementioned method, and the repetitions will not be repeated. The terms "unit" or "module" used below can be a combination of software and / or hardware that implements predetermined functions. Although the system described in the following embodiments is preferably implemented in software, hardware and a combination of software and hardware are also possible and conceived.

[0044] Specifically, Figure 2 is a schematic diagram of the module structure of an embodiment of the slope stability prediction device provided in this specification, such as Figure 2As shown, the slope stability prediction device provided in this specification includes: An acquisition module 201 is used to acquire training data and establish a PPO-PSO-BP prediction model for predicting slope stability; The training module 202 is used to input the training data into the PPO-PSO-BP prediction model, and train the particle swarm parameters according to the strategy network and the value network to obtain the optimal strategy π, and use the test data to iterate according to the optimal strategy to obtain the optimal particle. The particle swarm parameters specifically include the inertia weight , individual cognitive learning factors and social cognitive learning factors , the optimal strategy is specifically that the Actor network in the PPO algorithm outputs a probability to determine the movement mode of the particle in the search space, and the optimal particle is specifically the optimal weight parameter and the optimal bias parameter of the BP network in the PPO-PSO-BP prediction model; The prediction module 203 is used to input the slope data at the current moment into the PPO-PSO-BP prediction model after the parameters are determined to obtain the slope stability at the prediction moment.

[0045] It should be noted that the above-mentioned system may also include other implementation modes according to the description of the corresponding method embodiment. The specific implementation modes may refer to the description of the above-mentioned corresponding method embodiment, which will not be described one by one here.

[0046] The present application also provides an electronic device, including: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the method provided in the above embodiment.

[0047] The electronic device provided in the embodiment of the present application stores executable instructions of the processor in the memory. When the processor executes the executable instructions, it can first obtain training data and establish a PPO-PSO-BP prediction model for predicting slope stability; then input the training data into the PPO-PSO-BP prediction model, and train and optimize the particle swarm parameters according to the strategy network and the value network to obtain the optimal strategy, and use the test data to iterate according to the optimal strategy to obtain the optimal particle, and the particle swarm parameters specifically include inertia weight , individual cognitive learning factors and social cognitive learning factors The optimal strategy is that the PPO algorithm Actor network outputs a probability distribution to determine the movement of particles in the search space. The optimal particles are the optimal weight parameters and the optimal bias parameters of the BP network in the PPO-PSO-BP prediction model. Finally, the slope data at the current moment is input into the PPO-PSO-BP prediction model after the parameters are determined to obtain the slope stability at the prediction moment, avoiding the problems of poor population diversity, difficulty in convergence and falling into local extreme values ​​in the prior art, which can effectively improve the prediction accuracy and prediction stability of slope stability.

[0048] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0049] The methods or devices described in the above embodiments provided in this specification can implement business logic through computer programs and record them on storage media, and the storage media can be read and executed by computers to achieve the effects of the solutions described in the embodiments of this specification, such as: Obtain training data and establish a PPO-PSO-BP prediction model for predicting slope stability; The training data is input into the PPO-PSO-BP prediction model, and the particle swarm parameters are trained and optimized according to the strategy network and the value network to obtain the optimal strategy π. The optimal particles are iterated according to the optimal strategy using the test data. The particle swarm parameters specifically include the inertia weight , individual cognitive learning factors and social cognitive learning factors , the optimal strategy is specifically that the Actor network in the PPO algorithm outputs a probability to determine the movement mode of the particle in the search space, and the optimal particle is specifically the optimal weight parameter and the optimal bias parameter of the BP network in the PPO-PSO-BP prediction model; The slope data at the current moment is input into the PPO-PSO-BP prediction model after the parameters are determined to obtain the slope stability at the prediction moment.

[0050] The storage medium may include a physical device for storing information, which is usually a medium that digitizes information and then stores it in an electrical, magnetic or optical manner. The storage medium may include: a device that stores information in an electrical energy manner, such as various memories, such as RAM, ROM, etc.; a device that stores information in a magnetic energy manner, such as a hard disk, a floppy disk, a magnetic tape, a magnetic core memory, a magnetic bubble memory, a USB flash drive; a device that stores information in an optical manner, such as a CD or a DVD. Of course, there are other readable storage media, such as quantum memory, graphene memory, etc.

[0051] The embodiments of this specification are not limited to complying with industry communication standards, standard computer resource data update and data storage rules, or the situations described in one or more embodiments of this specification. Certain industry standards or slightly modified implementation plans based on the implementation described in the custom method or embodiment can also achieve the same, equivalent or similar, or predictable implementation effects after deformation of the above-mentioned embodiments. The embodiments obtained by using these modified or deformed data acquisition, storage, judgment, processing methods, etc. can still fall within the scope of the optional implementation plans of the embodiments of this specification.

[0052] The controller can be implemented in any appropriate manner, for example, the controller can take the form of a computer-readable medium such as a microprocessor or processor and a computer-readable program code (such as software or firmware) that can be executed by the (micro)processor, a logic gate, a switch, an application specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, ATMEL AT91SAM, MICROCHIP PIC18F26K20, and SILICONE LABS C8051F320. The memory controller can also be implemented as part of the control logic of the memory. It is also known to those skilled in the art that in addition to implementing the controller in a purely computer-readable program code manner, the controller can be implemented in the form of a logic gate, a switch, an application specific integrated circuit, a programmable logic controller, and an embedded microcontroller by logically programming the method steps. Therefore, such a controller can be considered as a hardware component, and the devices included therein for implementing various functions can also be considered as structures within the hardware component. Or even, the devices for implementing various functions can be considered as both software modules for implementing the method and structures within the hardware component.

[0053] The device embodiments described above are merely illustrative. For example, the division of the units is merely a logical function division. There may be other divisions in actual implementation, such as multiple units or plug-ins may be combined or integrated into another system, or some features may be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed may be through some interface, indirect coupling or communication connection of the device or unit, which may be electrical, mechanical or other forms.

[0054] These computer program instructions may also be loaded onto a computer or other programmable resource data updating device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions executed on the computer or other programmable device for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0055] Each embodiment in this specification is described in a progressive manner, and the same and similar parts between the embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment. In the description of this specification, the description of the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of this specification. In this specification, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples without contradiction.

[0056] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A slope stability prediction method, characterized in that: The method comprises: Obtain training data and establish a PPO-PSO-BP prediction model for predicting slope stability; The training data is input into the PPO-PSO-BP prediction model, and the particle swarm parameters are trained and optimized according to the strategy network and the value network to obtain the optimal strategy π, and the optimal particles are iterated according to the optimal strategy using the test data. The particle swarm parameters specifically include the inertia weight , individual cognitive learning factors and social cognitive learning factors , the optimal strategy is specifically that the Actor network in the PPO algorithm outputs a probability to determine the movement mode of the particle in the search space, and the optimal particle is specifically the optimal weight parameter and the optimal bias parameter of the BP network in the PPO-PSO-BP prediction model; The slope data at the current moment is input into the PPO-PSO-BP prediction model after the parameters are determined to obtain the slope stability at the prediction moment.

2. The slope stability prediction method according to claim 1, characterized in that: The particle swarm parameters are trained and optimized according to the strategy network and the value network to obtain the optimal strategy , including: The current state of the agent corresponding to the PPO algorithm is determined according to the current position and current fitness value of each particle in the particle swarm. Specifically, it includes the variance of the fitness value at the current moment, the particle aggregation, the current iteration progress, and the difference between the optimal fitness value of the population at the current moment and the preset accuracy. The current state of the agent is specifically determined by the following formula: ； ； ； ； ； ； In the formula, for The variance of the fitness value at each moment is expressed as express, for Moment The fitness value of a particle, for time The mean fitness value of particles, for The particle aggregation degree at the moment is express, is the search space where the particle swarm is located, for Moment The position of a particle, for The position of the central particle of the particle group at time is the current iteration progress, is the current iteration number, is the preset maximum number of iterations, is the difference between the optimal fitness value of the population at the current moment and the preset accuracy, for The optimal fitness value of the population at the moment; The strategy network is used to output the probability distribution of actions according to the current state of the agent, and actions are sampled from it. After inputting into the operating environment (particle swarm parameter optimization problem), the next state of the agent is determined , and get an instant reward After completing one iteration; The action For a three-dimensional action, : ； ； ； The three-dimensional action specifically corresponds to the particle swarm parameters, namely the inertia weight , ( ω ∈[0.4,0.9]), individual cognitive learning factor and social cognitive learning factors ; The Reward Set to: ； in, They are: ； ； ； ,when ; In the formula, for The optimal fitness value of the population at the moment, for The optimal fitness value of the population at the moment, for The variance of the fitness value at each moment, for The variance of the fitness value at each moment; for The aggregation degree of particles at a given moment, for The aggregation degree of particles at a given moment; The speed and position of the particles in the particle swarm are updated according to the action. The speed and position of the particles are updated specifically by the following formula: ； ； In the formula, for Moment The speed of a particle, for Moment The speed of a particle, For the The dimension of a particle, and for A random number in the range, For the The best historical position of a particle, is the optimal position of all particles, for Moment The position of a particle, for Moment The position of a particle; The current state of the agent ,action , Instant Rewards and the next state of the agent Form a trajectory and store the trajectory in the experience buffer; When the number of iterations reaches the maximum number of iterations of the round or the minimum fitness of the particle swarm reaches the preset accuracy, the value network is updated, and the advantage function is determined according to all trajectories in the experience buffer area through the updated value network, and then the strategy network is updated through the advantage function to end the current round.

3. The slope stability prediction method according to claim 2, characterized in that: Specifically, the parameters of the value network are calculated through the following formula: To update: ； In the formula, for Parameters of the moment value network, is the update step size of the value network parameters, for The time series difference error value, is the gradient derivative, for The state value function at the moment; Updating the policy network by the advantage function is specifically to update the objective function of the policy network by the following formula : ； ； ； In the formula, is the maximum iteration step, is the current iteration step, is the truncation function, is the mean calculation function, is the ratio of the new and old strategies, is the advantage function, is the minimum value, Yes Truncation is performed to limit the magnitude of the policy update. Based on the ratio of new and old strategies With advantage function The product of reflects the advantage of the action under the current strategy and the impact of strategy changes. Yes After truncation, the advantage function Multiply to avoid training instability due to large changes in strategy. Yes and The minimum value between the two ensures that a relatively conservative and stable update amount is selected during the policy update process; is the discount factor, To control the weights of multi-step returns in advantage estimation, for The time difference error value under the moment decision step, is the time series difference error value before the maximum iteration step, For status The value estimation error of for When for status The estimated value of the cumulative reward after a certain number of steps, for The state value function at time , which means that Start expecting future accumulated rewards, for The state value function at the moment, which represents the state from Start expecting future accumulating rewards; The parameters of the policy network The update is performed using the following formula: ； In the formula, for The parameters of the policy network at the decision step, is the policy network parameter update step size, for The time difference error value under the moment decision step, is the gradient derivative, is a logarithmic function, For the parameter Determined policy function, expressed in state Take action probability.

4. A slope stability prediction device, characterized in that: The device comprises: An acquisition module is used to acquire training data and establish a PPO-PSO-BP prediction model for predicting slope stability; The training module is used to input the training data into the PPO-PSO-BP prediction model, and train the particle swarm parameters according to the strategy network and value network to obtain the optimal strategy. , the optimal particles are obtained by iteration according to the optimal strategy using the test data, and the particle swarm parameters specifically include the inertia weight , individual cognitive learning factors and social cognitive learning factors , the optimal strategy is specifically that the Actor network in the PPO algorithm outputs a probability to determine the movement mode of the particle in the search space, and the optimal particle is specifically the optimal weight parameter and the optimal bias parameter of the BP network in the PPO-PSO-BP prediction model; The prediction module is used to input the slope data at the current moment into the PPO-PSO-BP prediction model after the parameters are determined to obtain the slope stability at the prediction moment.

Images