Real-time prediction method and system for industrial processes based on recursive random configuration networks

Abstract

The present invention belongs to the technical field of complex system modeling in industrial processes, and discloses a real-time prediction method and system for industrial processes based on a recursive random configuration network, data preprocessing; recursive random configuration network initialization; random configuration weights and biases, and a new method for constructing candidate neurons in a reserve pool; supervised selection of candidate neurons according to inequality constraints; screening of candidate neurons, and determination of the optimal candidate neurons; adding the optimal candidate neurons in the reserve pool to the network model; calculation of network output weights in combination with target output and reserve pool output; based on a projection algorithm, online updating of the trained output weights according to real-time data, and calculation of real-time output. The present invention uses a projection algorithm to update the network output weights online, which can reduce the overall computational complexity of the algorithm while ensuring prediction accuracy, and is suitable for application scenarios with high real-time requirements.

Description

Technical Field

[0001] The present invention belongs to the technical field of complex system modeling in industrial processes, and in particular relates to a real-time prediction method and system for industrial processes based on a recursive random configuration network. Background Art

[0002] Complex systems in industrial processes have multivariable dynamic evolution behaviors, most of which show chaotic characteristics. For such complex dynamic systems, relevant variable information is often obtained through observed time series. However, in actual processes, it is often difficult to capture all the order delay information involved in the variables, and the input order at different times is constantly changing. It is very challenging to study such nonlinear dynamic systems with uncertain orders.

[0003] Through the above analysis, the problems and defects of the prior art are as follows:

[0004] (1) When the system input order is unknown, the traditional method requires pre-order identification. This process increases the amount of algorithm calculation and reduces the training speed of the model. It is not suitable for industrial processes with high real-time requirements.

[0005] (2) Recurrent neural networks can store some historical information because of the feedback connections between their neurons, and can be used to solve the problem of modeling nonlinear dynamic systems with uncertain orders. However, during the training process, traditional recurrent neural networks need to adjust the network connection weights and biases based on the error gradient descent method. Therefore, the training process is computationally intensive, time-consuming, and prone to falling into local optimality, resulting in low efficiency.

[0006] (3) There are also feedback connections between neurons in the echo state network. Compared with the recurrent neural network, the echo state network calculates the output weights through the least squares method. The input weights and the reserve pool connection weights are randomly given and no longer change. However, the echo state network has the disadvantages of blind selection of network structure, sensitive weight parameters, and inability to guarantee the global approximation properties of the model. In addition, it is unable to adaptively adjust the range of weights and biases according to the input data, which affects the prediction accuracy of the model.

[0007] In the industrial field, the above mentioned technical issues have the following impacts:

[0008] 1. Insufficient real-time processing capability: The input order of a dynamic system changes in real time. In many industrial processes with high real-time requirements, such as production line monitoring and equipment failure prediction, data needs to be processed quickly and efficiently to make decisions. However, order identification increases the amount of calculation, resulting in slow model training and failure to meet real-time requirements.

[0009] 2. Low prediction accuracy: Traditional recursive neural networks are prone to falling into local optimality, and echo state networks have problems with blind selection of network structure and sensitivity of weight parameters, which will affect the prediction accuracy of the model. In industrial application scenarios with high accuracy requirements, such as quality inspection and fault diagnosis, these problems will lead to quality problems or miss important fault warnings.

[0010] 3. Inefficiency: The training process of traditional recurrent neural networks requires a lot of computing resources. This not only increases the cost of industrial applications, but also limits the application of these technologies in large-scale industrial systems.

[0011] 4. Difficulty in parameter adjustment: Existing technologies cannot adaptively adjust the range of weights and biases based on input data, which means that manual parameter adjustment is required, which not only requires a high level of expertise but also makes the application of the model very difficult.

[0012] 5. Insufficient model stability: The global approximation properties of the model cannot be guaranteed, which means that when faced with some extreme or unknown situations, the performance of the model will fluctuate greatly, which is unacceptable for industrial systems with high safety requirements.

[0013] In general, these technical issues limit the widespread application of learning models in industrial fields to a certain extent. Summary of the invention

[0014] In view of the problems existing in the prior art, the present invention provides a method and system for real-time prediction of industrial processes based on a recursive random configuration network. The purpose is to solve the problem of modeling nonlinear dynamic systems with uncertain orders. The recursive random configuration network can use the stored historical information for modeling, and only input the data at the current moment, without obtaining all order delay variables. Under the condition that the root mean square error of the output of the neural network model is reduced, the inequality constraints of the random configuration algorithm are used to find suitable neurons as candidate neurons. Different from the original random connection of the reserve pool, the recursive random configuration network redistributes the connection weights of the newly added neurons, and only retains the network connection weights of the newly added neurons to themselves and the original neurons, and sets the network connection weights of the original neurons to the newly added neurons to zero, that is, no connection. After determining the newly added neurons, the output weights are calculated using the least squares method, and finally the maximum allowable number of neurons in the reserve pool and the maximum allowable output error are used to determine whether the network is built.

[0015] The present invention is implemented as follows: a real-time prediction method for industrial processes based on a recursive random configuration network, which performs dynamic modeling through a recursive random configuration network. The method first performs data preprocessing and network initialization, then randomly gives weights and biases to construct candidate neurons, and uses inequality constraints to select neurons to add to the reserve pool. This process is repeated to optimize the network structure. Then, the output weights are calculated by the least squares method, and the model performance is evaluated based on the root mean square error. Finally, the projection algorithm is used to update the trained model based on real-time data, thereby ensuring the high accuracy and good generalization ability of the model and meeting the needs of real-time dynamic modeling.

[0016] Further, including:

[0017] S1, data preprocessing;

[0018] S2, recursive random configuration network initialization;

[0019] S3, randomly configures weights and biases, and uses a new method to construct candidate neurons in the reserve pool;

[0020] S4, supervised selection of candidate neurons based on inequality constraints;

[0021] S5, screening candidate neurons and determining the best candidate neurons;

[0022] S6, adding the optimal reserve pool candidate neurons to the network model;

[0023] S7, combining the target output and the reserve pool output to calculate the network output weight;

[0024] S8, based on the projection algorithm, updates the trained output weights online according to real-time data and calculates the real-time output.

[0025] Furthermore, S1 specifically includes: given a set of time series data, input samples: {u(1),u(2),...,u(n max )}={(y(1),u(1)),(y(2),u(2)),…,(y(n max ),u(n max ))}, target output: T = {t(2), t(3), …, t(n max +1)}, where y and u are the system output and the controlled input respectively. No order identification is required. The output y(n+1) at the next moment is predicted only by the current moment input (y(n), u(n)). max is the number of samples; set the maximum allowable number of neurons in the reserve pool N max , the maximum expected output error allowable value ε, the maximum number of candidate neurons generated G max , input weight distribution parameter γ={λmin ,λ min +Δλ,...,λ max}.

[0026] Furthermore, S2 specifically includes: initializing the output error vector e0=T, the model output error scaling factor 0<r<1, assuming that there is already a reserve pool containing N neurons, and the network model at this time is:

[0027] x(n)=g(W in,N u(n)+W r,N x(n-1)+b N )

[0028] y(n)=W out (x(n),u(n))

[0029] Where W in,N , W r,N and W out is the input weight, the reservoir connection weight and the output weight matrix, b N is the bias matrix, x(n) is the state vector of the reservoir at time n, and g is the activation function.

[0030] Further, S3 specifically includes: from γ = {λ min ,λ min +Δλ,...,λ max}, randomly select the input weight W in,N+1 , the connection weight of the reserve pool is W r,N+1 and bias b N+1 , substitute the activation function x(n)=g(W in,N+1 u(n)+W r,N+1 x(n-1)+b N+1 ) to get G max Group candidate neurons

[0031] Further, S4 specifically includes: max Substitute the candidate neurons into the inequality constraints of the random configuration algorithm:

[0032]

[0033] Where m is the output dimension, and the non-negative real number sequence {μ N+1}satisfy And μ N+1 ≤(1-r), and select candidate neurons that satisfy the inequality constraints.

[0034] Further, S5 specifically includes: defining a set of variables Select the ones that satisfy the inequality constraints and make ξ N+1The largest candidate neuron is used as the optimal reserve pool candidate neuron.

[0035] Further, S6 specifically includes: adding the optimal reserve pool candidate neurons to the network model, and directly calculating the output weight W of the neural network model according to the target output and the reserve pool output out , and calculate the output root mean square error of the neural network model, use the root mean square error as the loss function, and update the output error of the model e0: = e N+1 And the number of neurons in the reserve pool N=N+1.

[0036] Further, S7 specifically includes: if the output root mean square error ||e0||2 of the current network model is greater than the maximum expected output error allowable value ε, and the number of neurons in the reserve pool N is less than the maximum allowable number of neurons in the reserve pool N max , then repeat S3 to S6; if the output root mean square error ||e0||2 of the current network model is less than the maximum expected output error tolerance ε, or the number of neurons in the reserve pool N is greater than the maximum allowable number of neurons in the reserve pool N max , the training ends and a network model that meets the constraints of random configuration theory is obtained.

[0037] Further, S8 specifically includes: based on the projection algorithm, updating the trained output weights online according to the real-time data, and calculating the real-time output:

[0038]

[0039] y(n)=W out (n)g(n-1)

[0040] Where 0<a<1, c>0 are two constants, g(n) is the output of the reserve pool at time n, and e(n) is the output error before correction.

[0041] Another object of the present invention is to provide an industrial process real-time prediction system based on a recursive random configuration network, comprising:

[0042] A preprocessing module, used for data preprocessing;

[0043] Initialization module, used for recursive random configuration network initialization;

[0044] Random configuration module, used to randomly configure weights and biases, and adopt a new method to construct candidate neurons in the reserve pool;

[0045] Neuron selection module, used to select candidate neurons in a supervised manner according to inequality constraints;

[0046] The optimal candidate determination module is used to screen candidate neurons and determine the optimal candidate neurons;

[0047] An optimal neuron adding module is used to add the optimal reserve pool candidate neurons to the network model;

[0048] Output weight calculation module, used to calculate the network output weight by combining the target output and the reserve pool output;

[0049] The online update module is used to update the trained output weights online according to real-time data based on the projection algorithm and calculate the real-time output.

[0050] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows:

[0051] First, the present invention provides a method and system for real-time prediction of industrial processes based on a recursive random configuration network. To solve the problem of unknown order of dynamic data, the present invention does not require order identification and uses historical information stored in a reserve pool for modeling, thereby effectively improving the calculation rate of the model. To solve the problem of parameter sensitivity and structural uncertainty faced by traditional recursive neural networks and echo state networks, the present invention randomly configures the input weights and biases of neurons in the reserve pool, and can adaptively adjust the range of weights and biases according to the input data, without the need for iterative updates through back propagation, thus avoiding the problems of falling into local optimality, gradient vanishing or explosion caused by updating network parameters using the gradient descent method, and determining the network structure during the configuration process. In theory, the global approximation properties of the model can be guaranteed, and zero-error approximation of the training data can be achieved. When applied to industrial big data modeling training or prediction, the training speed and prediction accuracy of the model can be effectively improved.

[0052] Second, the present invention provides a method and system for real-time prediction of industrial processes based on a recursive random configuration network, which uses a projection algorithm to update the network output weights online to respond to dynamic changes in the system. Combined with the model framework designed by the present invention, it is possible to reduce the overall computational complexity of the algorithm while ensuring the prediction accuracy, and is suitable for application scenarios with high real-time requirements, and has good application prospects in the fields of industrial artificial intelligence, smart medical care, smart transportation, and unmanned driving.

[0053] For dynamic data modeling problems, most people stay at proving the existence of the global approximation ability of the model, but do not start from the perspective of constructivity. The present invention provides a real-time prediction method and system for industrial processes based on a recursive random configuration network, which incrementally constructs a dynamic model with global approximation ability. Moreover, the model does not require order identification, does not require gradient descent to update weights and biases, and can adaptively adjust the range of weights and biases, which is suitable for dynamic modeling problems with uncertain orders.

[0054] Third, the real-time prediction method for industrial processes based on recursive random configuration networks provided by the present invention brings the following significant technological advances in industry:

[0055] 1. Improve real-time processing capabilities: No order identification is required. Modeling is performed through historical information stored in the network reserve pool, and a new reserve pool construction method is adopted. This can effectively solve the order uncertainty problem in the dynamic process and quickly obtain the model output, making this method more suitable for industrial processes with high real-time requirements, such as real-time monitoring and fault detection.

[0056] 2. Improve prediction accuracy: This method selects candidate neurons in a supervised manner, combines the target output and the reserve pool output to calculate the network output weight, and further improves the prediction accuracy of the model. It can bring better performance in areas such as quality control and fault prediction.

[0057] 3. Improve model efficiency: This method is a type of random learning algorithm. The input weights and biases are randomly given, and the output weights are obtained by the least squares method. There is no need to update the weights by gradient descent. It can reduce the amount of computation in the training process, save a lot of computing resources in traditional methods, improve efficiency, and make it possible to apply this technology in large-scale industrial systems.

[0058] 4. Realize adaptive adjustment of model parameters: Based on the projection algorithm, the trained output weights are updated online according to real-time data, which realizes adaptive adjustment of model parameters, reduces the difficulty of manual adjustment of parameters, and makes the application of the model more convenient.

[0059] 5. Enhanced model stability: The supervised selection of candidate neurons based on inequality constraints can ensure the global approximation properties of the model; the projection algorithm used can ensure the convergence of the updated output weights, which enhances the stability of the model in the face of extreme or unknown situations, which is of great significance for industrial systems with high safety requirements.

[0060] In general, this approach will promote the widespread application of deep learning models in industrial fields and bring significant technological progress. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments of the present invention. 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 creative work.

[0062] Figure 1 It is a flow chart of a method for real-time prediction of industrial processes based on a recursive random configuration network provided by an embodiment of the present invention;

[0063] Figure 2 It is a schematic diagram of the structure of an industrial process real-time prediction system based on a recursive random configuration network provided by an embodiment of the present invention;

[0064] Figure 3 The fitting effect of RSCN on MG dynamic data verifies that the supervised construction of the reserve pool by our proposed method can effectively improve the prediction performance of the model;

[0065] Figure 4 This is a graph of the test results of each method under different reserve pool sizes;

[0066] Figure 5 It is the fitting effect diagram of each method on the nonlinear system identification data;

[0067] Figure 6 This is a diagram showing the fitting effects of various methods on the butane concentration data at the bottom of the debutanizer tower. DETAILED DESCRIPTION

[0068] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0069] In view of the problems existing in the prior art, the present invention provides a method and system for real-time prediction of industrial processes based on a recursive random configuration network. The present invention is described in detail below in conjunction with the accompanying drawings.

[0070] Application Example 1: Dynamic Path Planning of Industrial Robots

[0071] Step 1: Data collection. Use sensors to collect data about the robot’s current position, target position, obstacle position, and other dynamic environmental factors in real time.

[0072] Step 2: Recursive random configuration network initialization. Initialize the network parameters according to the robot's motion characteristics and environmental characteristics, including the number of neurons in the reserve pool and the initial weights.

[0073] Step 3: Generation of candidate neurons: Based on the data collected in step 1, a series of candidate neurons are generated using a random method, which potentially represent different path planning strategies.

[0074] Step 4: Neuron selection: By comparing with the real-time environment data, the inequality constraints are used to select the neurons that can provide the best path planning.

[0075] Step 5: Network update: Integrate the selected neurons into the reserve pool and update the network model to reflect the new dynamic environment information.

[0076] Step 6: Path planning execution. Based on the updated network model, the optimal path is calculated and the robot's control system guides the robot's movement in real time.

[0077] Step 7: Model online update. As the environment continues to change, the model is updated online using the projection algorithm and the new data collected to ensure the real-time and accuracy of path planning.

[0078] Application Example 2: Predictive Maintenance System

[0079] Step 1: Sensor deployment: Install vibration sensors, temperature sensors, etc. on key equipment to monitor the equipment operating status in real time.

[0080] Step 2: Network initialization: Initialize the parameters of the recursive random configuration network based on the historical maintenance data and operating characteristics of the device.

[0081] Step 3: Candidate neuron generation: Generate candidate neurons based on sensor data, which represent the potential failure modes and operating conditions of the equipment.

[0082] Step 4: Optimize the network structure. Use inequality constraints to select neurons that match the actual operating status of the equipment in order to more accurately identify potential faults.

[0083] Step 5: Network training. The selected neurons are added to the reserve pool, the output weights are updated using the least squares method, and all sensor data are integrated for network training.

[0084] Step 6: Fault prediction: Use the trained network model to predict the future operating status and fault time of the equipment, and make maintenance plans in advance.

[0085] Step 7: Model adaptive update: As the equipment runs and new data accumulates, the model is updated using online learning methods to ensure the timeliness and accuracy of the prediction results.

[0086] In both embodiments, the ability of recursive random configuration networks is used to handle and adapt to highly dynamic and uncertain environments, and its powerful real-time learning and adaptive capabilities have important application value in the fields of industrial automation and maintenance.

[0087] The present invention mainly improves the following problems and defects of the prior art and achieves significant technical progress:

[0088] Insufficient real-time data processing: Traditional industrial forecasting methods may be inefficient in processing real-time dynamic data and cannot respond to environmental changes in a timely manner. The present invention improves the efficiency of real-time data processing by adopting a recursive random configuration network.

[0089] Insufficient flexibility and adaptability of the model: Existing prediction models may lack sufficient flexibility and adaptability to adapt to changes in industrial processes. The present invention enhances the flexibility and adaptability of the model by randomly configuring weights and biases and selecting neurons according to inequality constraints.

[0090] Limitations of model performance evaluation and optimization: The existing technology may have limitations in model performance evaluation and optimization. The present invention updates the output weights by the least squares method and uses the root mean square error to evaluate the model performance, effectively optimizing the model.

[0091] The technical effects and significant technical progress brought about by the present invention in solving the problems of the prior art are as follows:

[0092] Improve the accuracy and speed of real-time prediction: Through optimized network initialization and weight configuration, as well as a real-time data update mechanism, the present invention significantly improves the accuracy and speed of real-time prediction of industrial processes.

[0093] Enhance the adaptability of the model: The use of recursive random configuration networks enables the model to better adapt to the dynamic changes of the industrial environment and improves the adaptability of the model.

[0094] Continuous optimization of model performance: By continuously screening and optimizing candidate neurons during operation, the present invention ensures continuous optimization and improvement of model performance.

[0095] Enhanced generalization capability of the model: Due to the adoption of real-time data update based on the projection algorithm, the model of the present invention can better process diverse data and enhance the generalization capability.

[0096] Providing effective model evaluation and optimization means: By combining the least square method and root mean square error evaluation, the present invention provides a more effective evaluation and optimization means for the model.

[0097] The essential technical feature of this technical solution is that dynamic modeling is performed through a recursive random configuration network. The method first performs data preprocessing and network initialization, then constructs candidate neurons by randomly selecting weights and biases, and uses inequality constraints to select neurons to add to the reserve pool. This process is repeated to optimize the network structure. Then the output weights are updated by the least squares method, and the model performance is evaluated based on the root mean square error. Finally, the projection algorithm is used to update the trained model based on real-time data, thereby ensuring the high accuracy and good generalization ability of the model to meet the needs of real-time dynamic modeling.

[0098] like Figure 1 As shown, the specific steps of the industrial process real-time prediction method based on recursive random configuration network provided by the embodiment of the present invention are as follows:

[0099] Step 1: Data preprocessing.

[0100] Step 2: Initialization of the recursive random configuration network.

[0101] Set the maximum number of neurons in the reserve pool N max , the initial number of neurons N = 0, the maximum expected output error allowable value ε, the distribution of input weights γ = {λ min ,λ min +Δλ,...,λ max}, the maximum number of candidate neurons generated G max , model output error initialization: e0=T.

[0102] Step 3: From γ = {λ min ,λ min +Δλ,...,λ max}, randomly select the input weight W in,N+1 , the connection weight of the reserve pool is W r,N+1 and bias b N+1 , substitute the activation function x(n)=g(W in,N+1 u(n)+W r,N+1 x(n-1)+b N+1 ) to get G max Group candidate neurons

[0103] Step 4. Settings G max Substitute the candidate neurons into the inequality constraints of the random configuration algorithm:

[0104]

[0105] Select candidate neurons that satisfy the inequality constraints.

[0106] Step 5. Define a set of variables: Substitute the candidate neurons that satisfy the inequality constraints into Select ξ N+1 The largest candidate neuron and retains the corresponding weight and bias W in,N+1 , W r,N+1 and b N+1 , proceed to the next step; otherwise, λ=λ+Δλ, when λ<λ max , return to step 3 to continue. If the above two conditions are not met, let τ∈(0,1-r), r=r+τ, λ=λ min , and then return to step 3 to continue. Add the best candidate neuron to the reserve pool to obtain the latest connection weights and biases and

[0107] Step 6: Update the neural network model, and use the least squares method to obtain the output weight of the model through the reserve pool output and the target output. Finally update the model output error e N =e N+1 And the number of neurons in the reserve pool N=N+1.

[0108] Step 7: When the output root mean square error of the model ||e0||2>ε, and N<N max , then return to step 3 to execute; otherwise, the neural network model training is completed.

[0109] Step 8: Based on the projection algorithm, the trained output weights are calculated according to the real-time data. Update online and calculate real-time output to analyze the generalization performance of the model.

[0110] like Figure 2 As shown, the industrial process real-time prediction system based on recursive random configuration network provided by the embodiment of the present invention includes:

[0111] A preprocessing module, used for data preprocessing;

[0112] Initialization module, used for recursive random configuration network initialization;

[0113] Random configuration module, used to randomly configure weights and biases, and adopt a new method to construct candidate neurons in the reserve pool;

[0114] Neuron selection module, used to select candidate neurons in a supervised manner according to inequality constraints;

[0115] The optimal candidate determination module is used to screen candidate neurons and determine the optimal candidate neurons;

[0116] An optimal neuron adding module is used to add the optimal reserve pool candidate neurons to the network model;

[0117] Output weight calculation module, used to calculate the network output weight by combining the target output and the reserve pool output;

[0118] The online update module is used to update the trained output weights online according to real-time data based on the projection algorithm and calculate the real-time output.

[0119] Preferably, the industrial process real-time prediction method and system based on recursive random configuration network provided by the embodiment of the present invention include:

[0120] S1, data preprocessing;

[0121] S2, recursive random configuration network initialization;

[0122] S3, randomly configures weights and biases, and uses a new method to construct candidate neurons in the reserve pool;

[0123] S4, supervised selection of candidate neurons based on inequality constraints;

[0124] S5, screening candidate neurons and determining the best candidate neurons;

[0125] S6, adding the optimal reserve pool candidate neurons to the network model;

[0126] S7, combining the target output and the reserve pool output to calculate the network output weight;

[0127] S8, based on the projection algorithm, updates the trained output weights online according to real-time data and calculates the real-time output.

[0128] S1 specifically includes: given a set of time series data, input samples: {u(1),u(2),...,u(n max )}={(y(1),u(1)),(y(2),u(2)),...,(y(n max ),u(n max ))}, target output: T = {t(2), t(3), ..., t(n max +1)}, where y and u are the system output and the controlled input respectively, and the output y(n+1) at the next moment is predicted only by the current moment input (y(n), u(n)), n max is the number of samples; set the maximum allowable number of neurons in the reserve pool N max , the maximum expected output error allowable value ε, the maximum number of candidate neurons generated G max , input weight distribution parameter γ={λ min ,λ min +Δλ,...,λ max}.

[0129] S2 specifically includes: initializing the output error vector e0=T, the model output error scaling factor 0<r<1, assuming that there is already a reserve pool containing N neurons, the network model at this time is:

[0130] x(n)=g(W in,N u(n)+W r,N x(n-1)+b N )

[0131] y(n)=W out (x(n),u(n))

[0132] Where W in,N , W r,N and W outis the input weight, the pool connection weight and the output weight, b N is the bias, x(n) is the state of the reservoir at time n, and g is the activation function.

[0133] S3 specifically includes: min ,λ min +Δλ,...,λ max}, randomly select the input weight W in,N+1 , the connection weight of the reserve pool is W r,N+1 and bias b N+1 , substitute the activation function x(n)=g(W in,N+1 u(n)+W r,N+1 x(n-1)+b N+1 ) to get G max Group candidate neurons

[0134] S4 specifically includes: max Substitute the candidate neurons into the inequality constraints of the random configuration algorithm:

[0135]

[0136] Where m is the output dimension, and the non-negative real number sequence {μ N+1}satisfy And μ N+1 ≤(1-r), and select candidate neurons that satisfy the inequality constraints.

[0137] S5 specifically includes: defining a set of variables Select the ones that satisfy the inequality constraints and make ξ N+1 The largest candidate neuron is used as the optimal reserve pool candidate neuron.

[0138] S6 specifically includes: adding the optimal reserve pool candidate neurons to the network model, and directly calculating the output weight W of the neural network model according to the target output and the reserve pool output. out , and calculate the output root mean square error of the neural network model, use the root mean square error as the loss function, and update the output error of the model e0: = e N+1 And the number of neurons in the reserve pool N=N+1.

[0139] S7 specifically includes: if the output root mean square error ||e0||2 of the current network model is greater than the maximum expected output error allowable value ε, and the number of neurons in the reserve pool N is less than the maximum allowable number of neurons in the reserve pool N max , then repeat S3 to S6; if the output root mean square error ||e0||2 of the current network model is less than the maximum expected output error tolerance ε, or the number of neurons in the reserve pool N is greater than the maximum allowable number of neurons in the reserve pool Nmax , the training ends and a network model that meets the constraints of random configuration theory is obtained.

[0140] S8 specifically includes: based on the projection algorithm, updating the trained output weights online according to the real-time data, and calculating the real-time output:

[0141]

[0142] y(n)=W out (n)g(n-1)

[0143] Where 0<a<1, c>0 are two constants, g(n) is the output of the reserve pool at time n, and e(n) is the output error before correction.

[0144] S3 to S6 are the key points of the present invention. S3 links the input data with the weight parameters and can adaptively adjust the weight distribution to accelerate error convergence. This data-dependent parameter selection method can effectively improve the prediction accuracy of the model.

[0145] S3 adopts a new reserve pool weight construction method. For the N+1th newly added neuron, only the network connection weight of the newly added neuron to the original neuron is retained. and its own network connection weight The connection weights of the original neurons to the newly added neurons are set to zero That is, no connection:

[0146]

[0147] in represents the connection weight of the ith neuron to the jth neuron, This construction method can ensure that when a neuron is added to the reserve pool, the output states of the first N reserve pool neurons do not change. It is only necessary to configure the new neurons that reduce the model training error to ensure the global approximation property of the network model.

[0148] The principle behind the inequality constraints of the random configuration algorithm mentioned in S4 is paraphrased as follows:

[0149] Assume that the vector space Γ on L2 space is dense and There exists 0<||g||<b g , Given 0<r<1 and a non-negative real number sequence {μ N+1},in μ N+1 ≤(1-r). Given the following formula:

[0150]

[0151] If the random basis function g N+1 The constructed output weight satisfy

[0152]

[0153] And the following inequality constraints are satisfied:

[0154]

[0155] but Where T is the target output, F N+1 It is the predicted output of the model with N+1 reserve pool neurons, that is, the constructed neural network model has a global approximation property.

[0156] S5 is to find the candidate neurons that can reduce the model training error the fastest, thereby accelerating the convergence speed.

[0157] The calculated output weights mentioned in S6 The specific principles are as follows:

[0158] Assume that the vector space Γ on L2 space is dense and exist Given 0<r<1 and a non-negative real number sequence μ N+1 ≤(1-r), and For the reserve pool neurons N = 1, 2, ..., given the following formula:

[0159]

[0160] If the random basis function g N+1 The constructed output weight W out,N+1 satisfy:

[0161]

[0162] And the following inequality constraints are satisfied:

[0163]

[0164] but

[0165]

[0166] but Right now The constructed neural network model has global approximation properties.

[0167] In addition, the suboptimal solution It can be obtained by the least squares method:

[0168]

[0169] Among them G N+1 is the matrix of random basis function g.

[0170] Taking the Mackey-Glass (MG) chaotic system and a nonlinear dynamic system as examples,

[0171] Example 1: The Mackey-Glass (MG) chaotic system is a very typical chaotic time series, which is generated by the following differential equation with time delay:

[0172]

[0173] Among them, when τ>16.8, the entire sequence is chaotic, has no period, and neither converges nor diverges.

[0174] Example 2: System identification is an important part of modern control theory and signal processing. The dynamic characteristics of the system are considered to be reflected in the changing input and output data. However, most actual systems are nonlinear, so nonlinear system identification becomes an important and complex problem. Given the following nonlinear system, in which u(n) = 1.05 × sin(n / 45) in the training stage, the mathematical model in the test stage is as follows:

[0175] y(n+1)=0.72y(n)+0.025y(n-1)u(n-1)+0.01u 2 (n-2)+0.2u(n-3)

[0176]

[0177] Example 3: The debutanizer process is an important part of the desulfurization and naphtha separation unit in the petroleum refining process. This process uses 7 auxiliary variables, namely: tower top temperature u1, tower top pressure u2, tower top reflux u3, tower top product outflow u4, sixth layer tower plate temperature u5, tower bottom temperature 1u6, tower bottom temperature 2u7, to predict the butane concentration y at the bottom of the tower. Its nonlinear model can be described as:

[0178] y(n)=f(u1(n),u2(n),u3(n),u4(n),u5(n),u5(n-1),u5(n-2),u5(n-3),(u6(n)+u7(n)) / 2,y(n-1),y(n-2),y(n-3),y(n-4))

[0179] Considering the unknown order, we reset the input of the above model, that is,

[0180] y(n)=f(u1(n),u2(n),u3(n),u4(n),u5(n),u6(n),u7(n),y(n-1))

[0181] In the experiment, we selected ESN, LSTM, SCR, PESN and LIESN to compare with the proposed method (RSCN). The Normalized root-mean square error (NRMSE) was selected as the evaluation index of the network prediction performance.

[0182] Example 1: Mackey-Glass (MG) chaotic system

[0183]

[0184] Analyzing the experimental results, LTSM has the largest prediction error, while RSCN has the smallest prediction error. In addition, compared with the original ESN, RSCN's prediction error is reduced by half. Figure 3 This is the fitting effect of RSCN on MG dynamic data, which verifies that our proposed method can construct the reserve pool in a supervised manner and effectively improve the prediction performance of the model.

[0185] Example 2: Nonlinear System Identification

[0186] Figure 4 The test results of each method under different reserve pool sizes and the fitting effect of each method are shown in Figure 2. Figure 5 It can be seen that when the reserve pool size exceeds 30, the prediction accuracy of RSCN is higher than that of other methods, and the prediction value of RSCN can effectively fit the target output, which verifies the effectiveness of the method proposed in this invention in nonlinear system identification.

[0187] Example 3: Soft measurement of butane concentration at the bottom of debutanizer

[0188] Figure 6 It is the prediction effect of each neural network model. It can be seen that the prediction error of RSCN is the smallest, and the fit between the model output and the target value is higher, which verifies the effectiveness of the method proposed in this invention in the soft measurement of industrial process parameters.

[0189] Application Example 1: Production Line Failure Prediction

[0190] In the production line, the prediction of equipment failure is very important, which can detect problems in advance, thereby reducing downtime and improving production efficiency. The real-time prediction method of industrial processes based on recursive random configuration networks can be applied here.

[0191] 1. Data preprocessing: Collect various sensor data from production line equipment, such as temperature, pressure, vibration, etc. Standardize or normalize the data for model training.

[0192] 2. Recursive random configuration network initialization: Initialize the network and set the number and connection method of neurons.

[0193] 3. Randomly configure weights and biases: Randomly assign weights and biases to construct candidate neurons in the reserve pool.

[0194] 4. Select candidate neurons: Select candidate neurons in a supervised manner based on the inequality constraints of the device operating status.

[0195] 5. Determine the optimal candidate neuron: By comparing the predicted results with the actual results, screen and determine the optimal candidate neuron.

[0196] 6. Calculate network output weight: Calculate network output weight based on the equipment operating status and reserve pool output.

[0197] 7. Update output weights online: Based on real-time data, use the projection algorithm to update the trained output weights online.

[0198] 8. Predict equipment failure: Based on the trained model, calculate the real-time output of the equipment and predict whether the equipment will fail.

[0199] Application Example 2: Energy System Load Forecasting

[0200] In energy systems, accurate load prediction is very important for energy saving and optimizing system operation. The real-time prediction method of industrial processes based on recursive random configuration networks can also be applied here.

[0201] 1. Data preprocessing: Collect the load data of the energy system and various factors that affect the load, such as weather conditions, holiday information, etc. Standardize or normalize the data for model training.

[0202] 2. Recursive random configuration network initialization: Initialize the network and set the number and connection method of neurons.

[0203] 3. Randomly configure weights and biases: Randomly assign weights and biases to construct candidate neurons in the reserve pool.

[0204] 4. Select candidate neurons: Select candidate neurons in a supervised manner based on the inequality constraints of the energy system load.

[0205] 5. Determine the optimal candidate neuron: By comparing the predicted results with the actual results, screen and determine the optimal candidate neuron.

[0206] 6. Calculate network output weight: Calculate network output weight by combining energy system load and reserve pool output.

[0207] 7. Update output weights online: Based on real-time data, use the projection algorithm to update the trained output weights online.

[0208] 8. Predict load: Based on the trained model, calculate the real-time output of the energy system and predict future load conditions.

[0209] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. It can be understood by a person of ordinary skill in the art that the above-mentioned devices and methods can be implemented using computer executable instructions and / or contained in a processor control code, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. Such code is provided on the carrier medium. The device and its modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, and can also be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0210] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principle of the present invention should be covered by the protection scope of the present invention.

Claims

1. A real-time prediction method for industrial processes based on a recursive random configuration network, characterized in that: Dynamic modeling is performed through a recursive random configuration network. The method first performs data preprocessing and network initialization, then constructs candidate neurons by randomly selecting weights and biases, and uses inequality constraints to select neurons to add to the reserve pool. This process is repeated to optimize the network structure; then the output weights are updated through the least squares method, and the model performance is evaluated based on the root mean square error. Finally, the projection algorithm is used to update the trained model based on real-time data; The method comprises the following steps: S1: Data preprocessing: Collect the load data of the energy system and various factors affecting the load; standardize or normalize the data for model training; S2: Recursive random configuration network initialization: Initialize the network and set the number and connection mode of neurons; S3: Randomly configure weights and biases: randomly assign weights and biases to construct candidate neurons in the reserve pool; S4: Select candidate neurons: Select candidate neurons in a supervised manner according to the inequality constraints of the energy system load; S5: Determine the best candidate neuron: Screen and determine the best candidate neuron by comparing the predicted results with the actual results; S6: Add the optimal reserve pool candidate neurons to the network model; S7: Calculate network output weight: Calculate network output weight by combining energy system load and reserve pool output; S8: Online update of output weights: Based on real-time data, the trained output weights are updated online using the projection algorithm; S9: Predicting load: Based on the trained model, calculate the real-time output of the energy system and predict future load conditions; S1 specifically includes: given a set of time series data, input sample: {u(1),u(2),…,u(n max )}={(y(1),u(1)),(y(2),u(2)),...,(y(n max ),u(n max ))}, target output: T = {t(2), t(3), ..., t(n max +1)}, where y and u are the system output and the controlled input respectively, and the output y(n+1) at the next moment is predicted only by the current moment input (y(n), u(n)), n max is the number of samples; set the maximum allowable number of neurons in the reserve pool N max , the maximum expected output error allowable value ε, the maximum number of candidate neurons generated G max , input weight distribution parameter γ={λ min ,λ min +Δλ,...,λ max }; S2 specifically includes: initializing the output error vector e0=T, assuming that there is already a reserve pool containing N neurons, and the network model at this time is: x(n)=g(W in,N u(n)+W r,N x(n-1)+b N ) y(n)=W out (x(n),u(n)) Where W in,N , W r,N and W out is the input weight, the pool connection weight and the output weight, b N is the bias, x(n) is the state of the reservoir at time n; S3 specifically includes: min ,λ min +Δλ,...,λ max }, randomly select the input weight W in,N+1 , the connection weight of the reserve pool is W r,N+1 and bias b N+1 , substitute into the activation function x(n)=g(W in,N+1 u(n)+W r,N+1 x(n-1)+b N+1 ) to get G max Group candidate neurons S4 specifically includes: max Substitute the candidate neurons into the inequality constraints of the random configuration algorithm: Step S5 specifically includes: defining a set of variables Select the ones that satisfy the inequality constraints and make ξ N+1 The largest candidate neuron is used as the optimal reserve pool candidate neuron; Model output error scaling factor 0<r<1, g is the activation function, e0:=e N+1 is the output error, where m is the output dimension, and the non-negative real number sequence {μ N+1 }satisfy And μ N+1 ≤(1-r), and select candidate neurons that satisfy the inequality constraints.

2. The industrial process real-time prediction method based on recursive random configuration network according to claim 1, characterized in that: S7 specifically includes: adding the optimal reserve pool candidate neurons to the network model, and directly calculating the output weight W of the neural network model according to the target output and the reserve pool output. out , and calculate the output root mean square error of the neural network model, use the root mean square error as the loss function, and update the output error of the model e0: = e N+1 And the number of neurons in the reserve pool N=N+1.

3. The industrial process real-time prediction method based on recursive random configuration network according to claim 1, characterized in that: S8 specifically includes: if the output root mean square error ||e0||2 of the current network model is greater than the maximum expected output error tolerance ε, and the number of neurons in the reserve pool N is less than the maximum allowable number of neurons in the reserve pool N max , then repeat S3 to S7; if the output root mean square error of the current network model ||e0||2 is less than the maximum expected output error tolerance ε, or the number of neurons in the reserve pool N is greater than the maximum allowable number of neurons in the reserve pool N max , the training ends and a network model that satisfies the constraints of random configuration theory is obtained; S9 specifically includes: based on the projection algorithm, online updating of the trained output weights according to real-time data, and calculation of real-time output: y(n)=W out (n)g(n-1) Where 0<a<1, c>0 are two constants, g(n) is the output of the reserve pool at time n, and e(n) is the output error before correction.

4. A real-time prediction system for industrial processes based on a recursive random configuration network using the prediction method according to any one of claims 1 to 3, characterized in that: include: A preprocessing module, used for data preprocessing; Initialization module, used for recursive random configuration network initialization; Random configuration module, used to randomly configure weights and biases, and adopt a new method to construct candidate neurons in the reserve pool; Neuron selection module, used to select candidate neurons in a supervised manner according to inequality constraints; The optimal candidate determination module is used to screen candidate neurons and determine the optimal candidate neurons; An optimal neuron adding module is used to add the optimal reserve pool candidate neurons to the network model; Output weight calculation module, used to calculate the network output weight by combining the target output and the reserve pool output; The online update module is used to update the trained output weights online according to real-time data based on the projection algorithm and calculate the real-time output.

Images