Methods, products, media and equipment for time series prediction of marine disaster-causing organisms in coastal nuclear power plants

Abstract

The present invention discloses a method, product, medium and equipment for predicting marine disaster-causing organisms in coastal nuclear power plants, and relates to the field of marine monitoring. First, the individual size and abundance time series data of target disaster-causing organisms in the target sea area of ​​cold source water intake of coastal nuclear power plants and the synchronous marine environmental element time series data are obtained, and the marine environmental influencing element time series data that affect the individual size and number proliferation of target disaster-causing organisms are screened out; the individual size and abundance time series prediction model is established by using the CRBM-DBN time series prediction algorithm; the hyperparameters of the time series prediction model are optimized by using the adaptive particle swarm optimization algorithm; the time series prediction model after the hyperparameter optimization is trained, and the trained time series prediction model is connected to the marine disaster-causing organism monitoring and early warning system, and the individual size and abundance of target disaster-causing organisms are predicted in real time, which can predict the individual size and abundance of disaster-causing organisms and their changing trends in the future time period in advance at the early stage of the development of disaster-causing organisms, thereby improving the prediction accuracy.

Description

Technical Field

[0001] The present invention relates to the field of marine monitoring technology, and in particular to a method, product, medium and equipment for predicting the time series of marine disaster-causing organisms in a coastal nuclear power plant. Background Art

[0002] Typical marine biohazards such as large algae, aquatic plants, jellyfish, pen snails, and shrimps will clog the cooling water intake of nuclear power plants, causing insufficient cooling water intake, leading to reduced power operation of the units or even shutdown events, posing a huge threat to the cooling source safety of coastal nuclear power plants and causing huge economic losses.

[0003] Early nuclear power plant response measures mainly included passive defense such as backwashing the cold source water intake filter drum net, monitoring the pressure difference on both sides of the drum net, and setting up pollution interception nets. This passive defense cannot win enough response time and is not ideal for reducing shutdown accidents. Therefore, nuclear power plants have gradually begun to develop marine disaster-causing biological monitoring and early warning systems, monitoring and identifying disaster-causing organisms through satellite remote sensing, photography, sonar and other means, and then using neural networks, deep learning and other algorithms to establish disaster-causing material quantity and location prediction methods, trying to issue early warnings before disaster-causing organisms reach the grid and filter drum net. At this time, the prediction time and prediction accuracy of the disaster-causing biological mass prediction method become an important link in determining whether the entire system can achieve effective forecasting.

[0004] In recent years, oceanographers have attempted to design a prediction model for marine biomass changes at nuclear power plant water intakes based on neural networks and deep learning algorithms. For example, Chinese patent publication number CN109934419A discloses a method for predicting marine biomass changes at nuclear power plant water intakes, including extracting features from auxiliary variable data that affect the number of marine organisms using thermometers, salinity meters, and oxygen meters, measuring marine biological data using sonar, and using neural network algorithms and deep learning algorithms for learning and training to obtain a biomass prediction model. This prediction method lacks the ability to predict time series data and can only obtain the current number of disaster-causing organisms through various environmental factors. It is unable to predict the number of disaster-causing organisms and their changing trends in advance, and therefore cannot serve as an effective disaster warning. For another example, Chinese patent CN1100991730A discloses a method for quantitatively predicting marine hazards for coastal nuclear power plants based on neural networks, including weight analysis of factors affecting marine hazards for coastal nuclear power plants to form a quantitative prediction index system for marine hazards for coastal nuclear power plants, collection of sample data of marine disasters that have occurred in coastal nuclear power plants, construction of a neural network function, and training to obtain a quantitative prediction model for marine hazards for coastal nuclear power plants; this prediction method only collects sample data of disasters that have occurred, and can only reflect the characteristics during the period when the disaster occurs. The trained model can only simulate the disaster situation that has occurred, and cannot achieve early prediction in the larval stage of disaster-causing organisms.

[0005] In general, the current prediction method for nuclear power cooling source disaster-causing organisms needs to be improved urgently. It mainly has two shortcomings. First, it only considers the number of disaster-causing organisms, ignoring that the development degree of the individual size of disaster-causing organisms is closely related to the proliferation of their biological population, which restricts the prediction accuracy of the model. When the size of the disaster-causing organisms is small enough to block the sewage net, even if the number is large, it is not enough to cause a disaster. The situation of inaccurate warning will occur due to only considering the amount of disaster-causing organisms; even when the individual size of the disaster-causing organisms has developed to the point where it can block the drum net, the disaster-causing organisms proliferate rapidly, and the time left for the nuclear power plant to react by predicting the number will be greatly reduced; therefore, it is very necessary to fully consider the individual size of the disaster-causing organisms to achieve early prediction and improve the prediction ability; second, the current prediction model lacks time series prediction ability, and can only obtain the current number of disaster-causing organisms by training various environmental factors. It is impossible to predict the number of disaster-causing organisms and their changing trends in advance, and it cannot play an effective role in early warning of disasters. Summary of the invention

[0006] The purpose of the present invention is to provide a method, product, medium and equipment for time series prediction of marine hazard-causing organisms in coastal nuclear power plants, which can predict the individual size and abundance of hazard-causing organisms and their changing trends in future time periods in advance at the early stage of their development, increase the time series prediction capability while taking into account the individual size and abundance characteristics of hazard-causing organisms, and improve the prediction accuracy.

[0007] To achieve the above object, the present invention provides the following solution.

[0008] In one aspect, the present invention provides a method for predicting the time series of marine disaster-causing organisms in a coastal nuclear power plant, comprising:

[0009] Obtain individual size time series data and abundance time series data of target disaster-causing organisms in the target sea area for cold source water intake of coastal nuclear power plants, as well as synchronized time series data of marine environmental elements; the target disaster-causing organisms include algae, water grasses, jellyfish, pen snails and hair shrimps;

[0010] Screen out the marine environment factor time series data that affect the individual size and number proliferation of target disaster-causing organisms from the marine environment factor time series data, including individual size marine environment factor time series data and abundance marine environment factor time series data;

[0011] The CRBM-DBN time series prediction algorithm was used to establish the individual size time series prediction model and abundance time series prediction model of the target disaster-causing organisms.

[0012] Adaptive particle swarm optimization algorithm is used to optimize the hyperparameters in the individual size time series prediction model and the abundance time series prediction model, and the hyperparameter-optimized individual size time series prediction model and the abundance time series prediction model are obtained.

[0013] The individual size time series data and the individual size marine environment impact factor time series data are used to train the individual size time series prediction model after hyperparameter optimization to obtain a trained individual size time series prediction model;

[0014] The individual size time series data, abundance time series data and abundance marine environment influencing factor time series data are used to train the abundance time series prediction model after hyperparameter optimization to obtain a trained abundance time series prediction model;

[0015] The trained individual size time series prediction model and abundance time series prediction model are connected to the marine disaster-causing organism monitoring and early warning system to make real-time predictions on the individual size and abundance of target disaster-causing organisms.

[0016] Optionally, the obtaining of individual size time series data and abundance time series data of target disaster-causing organisms in the target sea area for cold source water intake of a coastal nuclear power plant and synchronized marine environmental element time series data specifically includes:

[0017] Underwater microscopic imaging was used to conduct microscopic morphological imaging of all target disaster-causing organisms distributed in all water depths in the target sea area for cold source water intake of Binhai Nuclear Power Plant, and to obtain continuous long-term time series of individual size and abundance time series data of the target disaster-causing organisms from larvae to adults, reproduction, disaster and extinction throughout their life cycle;

[0018] An underwater bottom-profile stereoscopic observation platform method is used to obtain continuous long-term series of marine environmental element time series data covering all water depths that are synchronized with individual size time series data and abundance time series data; the marine environmental elements include temperature, salinity, pressure, dissolved oxygen, pH, chlorophyll, flow velocity, flow direction and wave height.

[0019] Optionally, the step of selecting the marine environment impact factor time series data that affects the individual size and number proliferation of target disaster-causing organisms from the marine environment factor time series data specifically includes:

[0020] The Pearson correlation coefficient method was used to establish correlation coefficient equations between the individual size time series data and abundance time series data of the target hazard-causing organisms and the time series data of each marine environmental element. At the same time, the day and night vertical movement characteristics of the target hazard-causing organisms were taken into consideration, and the correlation coefficient was solved after taking the average of the data of each water depth of the target hazard-causing organisms. The time series data of marine environmental elements that were strongly correlated with the individual size of the target hazard-causing organisms were determined as the time series data of marine environmental impact elements of individual size, and the time series data of environmental impact elements that were strongly correlated with the abundance of the target hazard-causing organisms were determined as the time series data of marine environmental impact elements of abundance.

[0021] Optionally, the use of the CRBM-DBN time series prediction algorithm to establish a time series prediction model for the individual size and a time series prediction model for the abundance of target disaster-causing organisms respectively includes:

[0022] Construct the CRBM-DBN time series prediction algorithm model based on the CRBM algorithm and DBN model;

[0023] The individual size time series data of the current time period and the individual size marine environment impact factor time series data are used as the input of the CRBM-DBN time series prediction algorithm model, and the individual size time series data of the future time period are used as the output of the CRBM-DBN time series prediction algorithm model to construct the individual size time series prediction model of the target disaster-causing organisms;

[0024] The individual size time series data, abundance time series data and abundance marine environment influencing factors time series data of the current time period are used as the input of the CRBM-DBN time series prediction algorithm model, and the abundance time series data of the future time period are used as the output of the CRBM-DBN time series prediction algorithm model to construct the abundance time series prediction model of the target disaster-causing organisms.

[0025] Optionally, the adaptive particle swarm optimization algorithm is used to optimize the hyperparameters in the individual size time series prediction model and the abundance time series prediction model respectively to obtain the individual size time series prediction model and the abundance time series prediction model after the hyperparameters are optimized, specifically including:

[0026] Initialize the particle population and set each particle to a multidimensional vector including the learning rate, the number of hidden layer nodes, and the number of hidden layer layers;

[0027] The reconstruction error of the individual size time series prediction model or the abundance time series prediction model is used as the fitness function, the model hyperparameters are trained by calculating the fitness of each particle, and the individual optimal particle and the global optimal particle with the maximum fitness are searched in the historical iterations;

[0028] After the training of the adaptive particle swarm optimization algorithm is completed, the optimal particles are selected to design the hyperparameters of the individual size time series prediction model or the abundance time series prediction model, and the individual size time series prediction model or the abundance time series prediction model after hyperparameter optimization is obtained.

[0029] Optionally, the training of the individual size time series prediction model after the hyperparameter optimization specifically includes:

[0030] The model is pre-trained using the contrastive divergence method to obtain better connection weights and biases for the input layer and hidden layer;

[0031] The BP algorithm is used to reverse tune the weights and biases obtained through pre-training to obtain the final parameters of the individual size time series prediction model.

[0032] Optionally, after obtaining the trained individual size time series prediction model, the method further includes:

[0033] The error evaluation function is used to evaluate the trained individual size time series prediction model. The smaller the error, the higher the prediction accuracy.

[0034] On the other hand, the present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements the method for predicting the time series of marine disaster-causing organisms in coastal nuclear power plants.

[0035] On the other hand, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for predicting the time series of marine disaster-causing organisms in coastal nuclear power plants.

[0036] On the other hand, the present invention also provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for predicting the time series of marine disaster-causing organisms in coastal nuclear power plants.

[0037] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0038] The present invention discloses a method, product, medium and equipment for predicting marine disaster-causing organisms in coastal nuclear power plants, which fully considers the influence of the development of the size of the individual of the disaster-causing organisms throughout their life cycle on the reproduction of the disaster-causing organisms, avoids the situation that the existing prediction method is inaccurate in warning when the number of disaster-causing organisms is large but the size is small enough to block the pollution interception net, improves the prediction accuracy, and realizes early prediction in the larval stage of the disaster-causing organisms. The present invention adopts the CRBM-DBN time series prediction algorithm, combines the characteristics of the continuous monitoring data of the individual size, abundance and synchronous marine environmental elements of the long series of disaster-causing organisms, fully considers the time series correlation of the individual size and abundance data of the disaster-causing organisms in a long range, and can predict the individual size and abundance of the disaster-causing organisms and their changing trends in the future time period in advance at the early stage of the development of the disaster-causing organisms, which can gain more reaction time for the nuclear power plant. The successful application of the method of the present invention can better meet the early prediction needs of biological disasters at the water intake of the coastal nuclear power plant. Furthermore, the present invention adopts an adaptive particle swarm optimization algorithm to optimize the hyperparameters of the time series prediction model, which solves the problems of high training cost and low accuracy caused by the traditional disaster-causing organism prediction method selecting structural parameters based on experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0040] Figure 1 A flow chart of a method for predicting the time series of marine disaster-causing organisms in a coastal nuclear power plant provided by the present invention;

[0041] Figure 2 A schematic diagram showing the principle of the method for predicting the time series of marine disaster-causing organisms in a coastal nuclear power plant provided by the present invention;

[0042] Figure 3 It is a schematic diagram of the CRBM-DBN time series prediction algorithm model of the present invention;

[0043] Figure 4 It is a structural diagram of the CRBM algorithm of the present invention. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] The present invention aims to provide a method, product, medium and equipment for time series prediction of marine hazard-causing organisms in coastal nuclear power plants, which can predict the individual size and abundance of hazard-causing organisms and their changing trends in future time periods in advance at the early stage of their development. While taking into account the individual size and abundance characteristics of hazard-causing organisms, the time series prediction capability is increased, thereby improving the prediction accuracy.

[0046] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0047] See also Figure 1 and Figure 2 The present invention provides a method for predicting the time series of marine disaster-causing organisms in coastal nuclear power plants, comprising:

[0048] Step 1: Obtain the individual size time series data and abundance time series data of target disaster-causing organisms in the target sea area for cold source water intake of coastal nuclear power plants, as well as the synchronized time series data of marine environmental elements.

[0049] The target disaster-causing organisms studied in this invention are typical marine disaster-causing organisms such as large algae, water grasses, jellyfish, pointed pen cap snails, and hair shrimps at the cold source water intake of the coastal nuclear power plant. The individual size and abundance time series data of the target disaster-causing organisms in the target sea area over the entire life cycle and the synchronized marine environment time series data are obtained online. The time series data of the disaster-causing organisms refers to the continuous time series monitoring data of all disaster-causing organisms distributed in various water depths from larvae to adult reproduction and then extinction.

[0050] Specifically, underwater microscopic imaging is used to perform microscopic morphological imaging of target disaster-causing organisms to obtain time series data of at least individual size and abundance of disaster-causing organisms. Abundance refers to the ratio of the number of target disaster-causing organisms in a specified time period to the water body photographed in that time period.

[0051] Specifically, the underwater bottom-seated profile stereo observation platform and other methods are used to obtain the time series data of marine environmental elements in real time, including at least temperature, salinity, pressure, dissolved oxygen, pH, chlorophyll, flow velocity, flow direction, wave height and other marine environmental elements, which are synchronized with the time series data of the size / abundance of the individual disaster-causing organisms and are continuous time series monitoring data covering all water depths. The symbol " / " is used in the present invention to represent "or".

[0052] Step 2: Filter out the marine environmental impact factor time series data that affect the individual size and number proliferation of target disaster-causing organisms from the marine environmental factor time series data, including the individual size marine environmental impact factor time series data and abundance marine environmental impact factor time series data.

[0053] The Pearson correlation coefficient method was used to establish the correlation coefficient equations between the time series data of the individual size / abundance of target hazard-causing organisms and the time series data of various marine environmental elements. At the same time, considering the characteristics of the vertical movement of hazard-causing organisms during the day and night, the average value of the multi-water depth time series data of the three-dimensional profile of hazard-causing organisms was taken to solve the correlation. All strongly correlated marine environmental elements were determined as the marine environmental influencing factor group that affects the individual size and number proliferation of hazard-causing organisms.

[0054] Specifically, the marine environment impact factor group of the target disaster-causing organisms includes an individual size marine environment impact factor group and an abundance marine environment impact factor group, wherein the individual size marine environment impact factor group includes all marine environment factors that are strongly related to the individual size of the disaster-causing organisms, and the abundance marine environment impact factor group includes all marine environment factors that are strongly related to the abundance of the disaster-causing organisms.

[0055] Specifically, the Pearson correlation coefficient method is used to calculate the correlation coefficient equation between the individual size / abundance of target disaster-causing organisms and marine environmental factors:

[0056]

[0057] Among them, R a It is the correlation coefficient between the individual size / abundance of target hazard-causing organisms and marine environmental factors considering the characteristics of vertical movement of hazard-causing organisms during the day and night, where a = 1, 2; R1 is the correlation coefficient between individual size and each marine environmental factor, and R2 is the correlation coefficient between abundance and each marine environmental factor. Cov is the covariance, σ is the standard deviation, and E is the expectation. X d Y is the average value of the time series data of marine environmental elements in all water depth layers of a certain marine environmental element profile; dIt is the average of the individual size or abundance time series data of all water depth layers of the target disaster-causing biological profile. A correlation coefficient value greater than 0.5 indicates strong correlation, 0.3-0.5 indicates moderate correlation, 0.1-0.3 indicates weak correlation, and less than 0.1 indicates extremely weak correlation.

[0058] When calculating, take one of the time series data of ocean environment elements such as temperature, salinity, pressure, dissolved oxygen, pH, chlorophyll, current velocity, current direction, wave height, etc. and find the average value of all water depths and substitute it into X d , take one of the individual size or abundance time series data and find the average value of all water depths and substitute it into Y d To calculate R a For example, X d is the average value of temperature time series data, Y d is the average value of individual size time series data. If the calculated R a If it is 0.6, it means that temperature is strongly correlated with the individual size of disaster-causing organisms, and temperature is recorded as one of the factors affecting the marine environment.

[0059] Step 3: Use the CRBM-DBN time series prediction algorithm to establish the individual size time series prediction model and abundance time series prediction model of the target disaster-causing organisms.

[0060] Based on the CRBM (Conditional restricted Boltzmann machines) algorithm and the DBN (deep belief network) model, the CRBM-DBN time series prediction algorithm model is constructed. Its network structure is as follows: Figure 3 shown.

[0061] The individual size time series data of the current time period and the synchronized individual size time series data of marine environmental impact factors are used as the input of the CRBM-DBN time series prediction algorithm model, and the individual size time series data of the future time period is used as the output of the CRBM-DBN time series prediction algorithm model to construct the individual size time series prediction model of the target disaster-causing organisms.

[0062] The individual size time series data, abundance time series data and synchronized abundance marine environment influencing factors time series data of the current time period are used as the input of the CRBM-DBN time series prediction algorithm model, and the abundance time series data of the future time period are used as the output of the CRBM-DBN time series prediction algorithm model to construct the abundance time series prediction model of the target disaster-causing organisms.

[0063] Among them, the input layer matrix of the individual size / abundance time series prediction model can be expressed as:

[0064]

[0065] Among them, x i,t is the value of the i-th input element in the input layer at time t. For the individual size time series prediction model, the n input elements refer to the marine environmental elements that are strongly correlated with the individual size of the hazard-causing organisms selected in step 2 plus the synchronized individual size of the hazard-causing organisms, a total of n parameters; and for the abundance time series prediction model, the n input elements refer to the marine environmental elements that are strongly correlated with the abundance of the hazard-causing organisms selected in step 2 plus the synchronized individual size and abundance parameters of the hazard-causing organisms. M represents the time window of the input data, and τ represents the time interval of the input data. M and τ can be flexibly adjusted according to the specific prediction time.

[0066] The individual size and abundance of the disaster-causing organisms in the future time period t+Mτ are taken as the output of the individual size and abundance time series prediction model, respectively, and the output layer matrix Y can be expressed as:

[0067] Y=[Y(t+τ),Y(t+2τ),…,Y(t+Mτ)] T (3)

[0068] The CRBM-DBN time series prediction algorithm model adopted in the present invention adds a short-term time series structure on the basis of the traditional deep belief network. Specifically, autoregressive weights are added between the input layer neurons and the weights between the historical input nodes and the current hidden layer nodes. The algorithm update is manifested in that a dynamic bias function is used instead of a static bias, which increases the learning of historical data rules and can extract more data features. While inheriting the reasoning process and training process of the traditional deep belief network, the algorithm's time series prediction capability is increased on the basis of the traditional deep belief network algorithm, thereby achieving better data time series prediction.

[0069] Step 4: Adaptive particle swarm optimization algorithm is used to optimize the hyperparameters in the individual size time series prediction model and the abundance time series prediction model, respectively, to obtain the individual size time series prediction model and the abundance time series prediction model after hyperparameter optimization.

[0070] The adaptive particle swarm optimization algorithm (APSO) is used to optimize the hyperparameters (mainly including the number of hidden layer nodes, the number of hidden layer layers and the learning rate) in the individual size / abundance time series prediction model of disaster-causing organisms. The specific process is as follows:

[0071] Step 4.1: Initialize the particle population as the initial group, set the number of particles N on the acceleration track, generally between 24 and 30. Each particle is set to a multidimensional vector including the learning rate, the number of nodes in the hidden layer, and the number of hidden layers.

[0072] Step 4.2: Use particle 1, particle 2, ... particle N to train model hyperparameters respectively;

[0073] Specifically, the reconstruction error of the CRBM-DBN time series prediction algorithm model (individual size / abundance time series prediction model) is used as the fitness function, and the fitness training model is calculated for each particle, and the particle with the maximum fitness is searched in the historical iteration. and the global best particle

[0074] Step 4.3: Update the velocity and position of each particle using equations (4) and (5);

[0075]

[0076] Where w, c1, c2 are the inertia weight and acceleration factor values ​​at the kth iteration respectively; r1, r2∈Rand[0,1], that is, random numbers between 0 and 1; are the speed, position, individual optimal position and population optimal position of the j-th dimension variable of parameter i in the k-th iteration respectively.

[0077] Step 4.4: When the fitness function reaches the preset value or the number of iterations reaches the preset value, the APSO algorithm training is terminated, and the optimal particles are selected to design the time series prediction model hyperparameters to obtain the hyperparameter-optimized individual size / abundance time series prediction model. Otherwise, go to step 4.3 to continue training the model.

[0078] Step 5: Use the individual size time series data and the individual size marine environment impact factor time series data to train the individual size time series prediction model after hyperparameter optimization to obtain a trained individual size time series prediction model.

[0079] The hyperparameters optimized in step 4 are assigned to the model and the model is trained. The accuracy is evaluated using the error evaluation function to obtain a better disaster-causing biological time series prediction model.

[0080] Specifically, the individual size time series data obtained in step 1 and the individual size marine environment impact factor time series data obtained in step 2 are Gaussian normalized as the input layer of the individual size prediction model. The energy function of the CRBM-DBN after Gaussian distribution processing is:

[0081]

[0082] In the above formula, v = (vi) and h = (h j ) are the vectors of the input layer units and hidden layer units of CRBM respectively; θ={w,a,b}={w=(wij)n × m, is the parameter of CRBM algorithm; w ij is the input layer unit v i and the hidden layer unit h j Symmetric connection weights between ; is the input layer unit v i The dynamic bias of is the hidden layer unit h j Dynamic bias of is the input layer unit v i The standard deviation of the Gaussian noise; n and m are the input layer units v i and the hidden layer unit h j The number of.

[0083] like Figure 4 As shown in the figure, the contrastive divergence method is used to pre-train CRBM layer by layer, and the state of the hidden nodes of CRBM is updated. The state of the jth hidden layer unit is:

[0084]

[0085] where σ(x) = 1 / (1+e -x ); λ is a random number between 0 and 1; N j (0,1) represents a Gaussian random variable with mean 0 and variance 1.

[0086] According to the state value of the hidden layer node, the state of the input layer node is updated again. The state of the i-th input layer unit is:

[0087]

[0088] The contrastive divergence method is used to pre-train the model to obtain better weights and biases. The dynamic bias function update formula of the CRBM network is:

[0089]

[0090] in, is the dynamic bias of input layer unit i, represents the dynamic bias of hidden layer unit j; a i and b j Represent the static bias of the original RBM algorithm for input layer unit i and hidden layer unit j respectively; is the state of the input layer unit k at time tq; is the weight value of the directed connection between input layer unit k and input layer unit i at time tq; is the weight value of the directed connection between the input layer unit k and the hidden layer unit j at time tq.

[0091] like Figure 3 and Figure 4As shown, after the contrast divergence method is used to train layer by layer to initialize the weights and bias of the entire network, the BP algorithm is used to reversely tune the connection weights and biases of the input layer and hidden layer obtained by pre-training to obtain the final individual size time series prediction model and abundance time series prediction model parameters. Figure 3 The network structure of DBN is retained, and the original RBM algorithm in DBN is replaced by the CRBM algorithm.

[0092] After the model training is completed, the error evaluation function is used for evaluation. The error evaluation function can be:

[0093]

[0094] Among them, n represents the length of data to be predicted, y is the true value of the i-th data, and y' is the predicted value of the i-th data. MAE is the average of the absolute values ​​of the deviations of all single observations from the arithmetic mean. RMSE is the root mean square error, which can well reflect the accuracy of the prediction. MRE uses the same unit dimension to reflect the degree to which the prediction result deviates from the true value. The three evaluate the prediction results from different angles, and all conform to the rule that the smaller the value, the higher the prediction accuracy.

[0095] Step 6: Use individual size time series data, abundance time series data and abundance marine environment influencing factor time series data to train the abundance time series prediction model after hyperparameter optimization to obtain a trained abundance time series prediction model.

[0096] Similar to the training process of the individual size time series prediction model, in the training process of the abundance time series prediction model, the contrast divergence method is used to pre-train the model to obtain the optimal connection weights and biases of the input layer and the hidden layer; then the BP algorithm is used to reversely tune the weights and biases obtained by pre-training to obtain the final abundance time series prediction model parameters. After the model training is completed, the error evaluation function (10) is used to evaluate the trained abundance time series prediction model. The smaller the error, the higher the prediction accuracy.

[0097] Step 7: Connect the trained individual size time series prediction model and abundance time series prediction model to the marine disaster-causing organism monitoring and early warning system to make real-time predictions on the individual size and abundance of target disaster-causing organisms.

[0098] The individual size and abundance time series prediction model obtained through training in step 6 is collectively referred to as the typical disaster-causing organism time series prediction model for coastal nuclear power plants. It is connected to the marine disaster-causing organism monitoring and early warning system for real-time prediction. The model prediction results are continuously compared in real time with the actual measured data of the monitoring and early warning system at each moment, and the model is verified and optimized according to the error situation.

[0099] On the one hand, the method of the present invention overcomes the shortcomings of existing prediction methods that only consider the number of disaster-causing organisms but ignore the influence of the development degree of the individual size of disaster-causing organisms, which restricts the prediction accuracy of the model; on the other hand, it avoids the situation that the existing prediction methods give inaccurate warnings when the number of disaster-causing organisms is large but the size is small enough to block the pollution-blocking net. The individual size / abundance time series prediction model trained by the present invention can be used to predict the individual size and abundance of disaster-causing organisms and their changing trends in the future time period in advance in the early stage of the development of disaster-causing organisms. Nuclear power plant staff can formulate accurate prevention and control plans based on the prediction results before the disaster occurs, so as to gain more reaction time for the nuclear power plant, and help nuclear power plant staff to formulate more detailed and accurate prevention and control plans. For example, according to the size and number characteristics of disaster-causing organisms in the waters near the nuclear power plant, staff can configure pollution-blocking nets according to different apertures and reasonably select the type of pollution-blocking nets. The successful application of the method of the present invention can better meet the early prediction needs of biological disasters at the water intakes of coastal nuclear power plants.

[0100] In some embodiments, the present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements the method for predicting the time series of marine disaster-causing organisms in coastal nuclear power plants.

[0101] In some embodiments, the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for predicting the timing of marine disaster-causing organisms in coastal nuclear power plants.

[0102] In some embodiments, the present invention also provides a computer device, which includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store pending transactions. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it can implement the marine disaster-causing biological time series prediction method of the coastal nuclear power plant.

[0103] The present invention provides a method for predicting marine disaster-causing organisms in coastal nuclear power plants taking into account the individual size of the disaster-causing organisms. The method obtains the individual size, abundance and continuous monitoring data of synchronous marine environmental elements of the continuous long-term and spatial sequence of disaster-causing organisms in nuclear power sea areas from larvae to adult reproduction and disaster to extinction in real time, and adopts the CRBM-DBN time series prediction algorithm to build a time series prediction model for disaster-causing organisms in nuclear power sea areas. The time series prediction ability of the CRBM-DBN algorithm is used to predict the individual size and abundance of disaster-causing organisms and their changing trends in the future time period in advance at the early stage of the development of the disaster-causing organisms, which can buy more reaction time for the nuclear power plant and help the nuclear power personnel to formulate more detailed and accurate prevention and control plans. For example, according to the size and quantity characteristics of the disaster-causing organisms in the waters near the nuclear power plant, the staff can configure the pollution interception net according to different apertures and reasonably select the type of the pollution interception net. The successful application of the method of the present invention can better meet the early prediction needs of biological disasters at the water intake of coastal nuclear power plants, and has broad application prospects.

[0104] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for predicting marine disaster-causing organisms in coastal nuclear power plants, characterized in that: include: Obtain the individual size time series data and abundance time series data of target disaster-causing organisms in the target sea area for cold water intake of coastal nuclear power plants, as well as the synchronized time series data of marine environmental elements; The target disaster-causing organisms include algae, water plants, jellyfish, pen snails and hair shrimps; Screen out the marine environment factor time series data that affect the individual size and number proliferation of target disaster-causing organisms from the marine environment factor time series data, including individual size marine environment factor time series data and abundance marine environment factor time series data; The CRBM-DBN time series prediction algorithm was used to establish the individual size time series prediction model and abundance time series prediction model of the target disaster-causing organisms. Adaptive particle swarm optimization algorithm is used to optimize the hyperparameters in the individual size time series prediction model and the abundance time series prediction model, and the hyperparameter-optimized individual size time series prediction model and the abundance time series prediction model are obtained. The individual size time series data and the individual size marine environment impact factor time series data are used to train the individual size time series prediction model after hyperparameter optimization to obtain a trained individual size time series prediction model; The individual size time series data, abundance time series data and abundance marine environment influencing factor time series data are used to train the abundance time series prediction model after hyperparameter optimization to obtain a trained abundance time series prediction model; The trained individual size time series prediction model and abundance time series prediction model are connected to the marine disaster-causing organism monitoring and early warning system to make real-time predictions on the individual size and abundance of target disaster-causing organisms.

2. The method for predicting marine disaster-causing organisms in coastal nuclear power plants according to claim 1, characterized in that: The acquisition of individual size time series data and abundance time series data of target disaster-causing organisms in the target sea area for cold source water intake of coastal nuclear power plants and synchronized marine environmental element time series data specifically includes: Underwater microscopic imaging was used to conduct microscopic morphological imaging of all target disaster-causing organisms distributed in all water depths in the target sea area for cold source water intake of Binhai Nuclear Power Plant, and to obtain continuous long-term time series of individual size and abundance time series data of the target disaster-causing organisms from larvae to adults, reproduction, disaster and extinction throughout their life cycle; An underwater bottom-profile stereoscopic observation platform method is used to obtain continuous long-term series of marine environmental element time series data covering all water depths that are synchronized with individual size time series data and abundance time series data; the marine environmental elements include temperature, salinity, pressure, dissolved oxygen, pH, chlorophyll, flow velocity, flow direction and wave height.

3. The method for predicting marine disaster-causing organisms in coastal nuclear power plants according to claim 1, characterized in that: The time series data of marine environmental impact factors that affect the size and number of target disaster-causing organisms are selected from the time series data of marine environmental factors, specifically including: The Pearson correlation coefficient method was used to establish correlation coefficient equations between the individual size time series data and abundance time series data of the target hazard-causing organisms and the time series data of each marine environmental element. At the same time, the day and night vertical movement characteristics of the target hazard-causing organisms were taken into consideration, and the correlation coefficient was solved after taking the average of the data of each water depth of the target hazard-causing organisms. The time series data of marine environmental elements that were strongly correlated with the individual size of the target hazard-causing organisms were determined as the time series data of marine environmental impact elements of individual size, and the time series data of environmental impact elements that were strongly correlated with the abundance of the target hazard-causing organisms were determined as the time series data of marine environmental impact elements of abundance.

4. The method for predicting marine disaster-causing organisms in coastal nuclear power plants according to claim 1, characterized in that: The CRBM-DBN time series prediction algorithm is used to respectively establish a time series prediction model for the individual size and a time series prediction model for the abundance of target disaster-causing organisms, specifically including: Construct the CRBM-DBN time series prediction algorithm model based on the CRBM algorithm and DBN model; The individual size time series data of the current time period and the individual size marine environment impact factor time series data are used as the input of the CRBM-DBN time series prediction algorithm model, and the individual size time series data of the future time period are used as the output of the CRBM-DBN time series prediction algorithm model to construct the individual size time series prediction model of the target disaster-causing organisms; The individual size time series data, abundance time series data and abundance marine environment influencing factors time series data of the current time period are used as the input of the CRBM-DBN time series prediction algorithm model, and the abundance time series data of the future time period are used as the output of the CRBM-DBN time series prediction algorithm model to construct the abundance time series prediction model of the target disaster-causing organisms.

5. The method for predicting marine disaster-causing organisms in coastal nuclear power plants according to claim 1, characterized in that: The method uses the adaptive particle swarm optimization algorithm to optimize the hyperparameters in the individual size time series prediction model and the abundance time series prediction model, respectively, to obtain the individual size time series prediction model and the abundance time series prediction model after the hyperparameters are optimized, specifically including: Initialize the particle population and set each particle to a multidimensional vector including the learning rate, the number of hidden layer nodes, and the number of hidden layer layers; The reconstruction error of the individual size time series prediction model or the abundance time series prediction model is used as the fitness function, the model hyperparameters are trained by calculating the fitness of each particle, and the individual optimal particle and the global optimal particle with the maximum fitness are searched in the historical iterations; After the training of the adaptive particle swarm optimization algorithm is completed, the optimal particles are selected to design the hyperparameters of the individual size time series prediction model or the abundance time series prediction model, and the individual size time series prediction model or the abundance time series prediction model after hyperparameter optimization is obtained.

6. The method for predicting marine disaster-causing organisms in coastal nuclear power plants according to claim 1, characterized in that: The training of the individual size time series prediction model after the hyperparameter optimization specifically includes: The model is pre-trained using the contrastive divergence method to obtain better connection weights and biases for the input layer and hidden layer; The BP algorithm is used to reverse tune the weights and biases obtained through pre-training to obtain the final parameters of the individual size time series prediction model.

7. The method for predicting marine disaster-causing organisms in coastal nuclear power plants according to claim 1, characterized in that: After obtaining the trained individual size time series prediction model, the method further includes: The error evaluation function is used to evaluate the trained individual size time series prediction model. The smaller the error, the higher the prediction accuracy.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for predicting the time series of marine disaster-causing organisms in a coastal nuclear power plant according to any one of claims 1 to 7 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for predicting the time series of marine disaster-causing organisms in a coastal nuclear power plant according to any one of claims 1 to 7 is implemented.

10. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for predicting marine disaster-causing organisms in coastal nuclear power plants according to any one of claims 1 to 7.

Images