Intelligent operation risk analysis method and system based on deep learning

Abstract

The invention discloses an intelligent operation risk analysis method and system based on deep learning, and relates to the technical field of power plant management, and the method comprises the steps: collecting the operation original data of power plant equipment in real time; carrying out data fusion on the collected data, and extracting equipment evaluation data; analyzing the equipment evaluation data through the long-short-term memory neural network model, and predicting an equipment fault based on an analysis result; based on the equipment fault prediction result, generating an equipment scheduling scheme by using a particle swarm optimization algorithm, and optimizing the scheduling scheme; and converting the optimized equipment scheduling scheme into a control instruction, and generating and executing equipment scheduling through a low-code development platform. According to the intelligent operation risk analysis method based on deep learning, a scheduling scheme can be optimized according to the health state and the fault risk of the equipment, the operation efficiency of the equipment is improved, a repair and maintenance plan is effectively arranged, and the problem of excessive maintenance or repair lag is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of power plant management, and in particular, to a method and system for intelligent operation risk analysis based on deep learning. Background Art

[0002] Currently, with the rapid development of intelligent and automated technologies, the operation management of power plant equipment faces increasingly complex challenges. Traditional power plant equipment scheduling methods mainly rely on manual experience and rule-based scheduling systems. Although this method can ensure the normal operation of equipment, it often lacks flexibility and real-time performance when facing equipment failures and emergencies. Traditional equipment failure prediction methods usually rely on historical data analysis. However, due to the complexity of data processing and modeling, the prediction accuracy is relatively low, and it is difficult to achieve early warning and effective prevention of failures. Existing methods have not effectively combined the failure prediction results with equipment scheduling dynamically, resulting in the inability of the equipment scheduling plan to make optimized decisions in real-time risk management, thereby affecting the operation efficiency of the equipment and the stability of the overall system.

[0003] In order to improve the operation efficiency of power plant equipment, many studies have begun to explore intelligent scheduling methods based on artificial intelligence and deep learning. However, most of these methods focus on the optimization of single tasks, such as only focusing on equipment scheduling or failure prediction, and lack a comprehensive solution that systematically combines failure prediction with equipment scheduling. In addition, although deep learning has shown good performance in failure prediction, for the health state assessment of equipment and the optimization of dynamic scheduling, traditional deep learning models mostly rely on static training data and lack the dynamic optimization ability based on real-time data, and are unable to effectively respond to changes in equipment states and the occurrence of sudden failures. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the technical problems solved by the present invention are as follows: Existing power plant equipment monitoring and failure prediction methods have problems such as relying on manual analysis, slow response speed, lack of flexibility, and insufficient failure prediction accuracy, making it difficult to cope with the dynamic changes in equipment operation states, and having limited optimization effects in equipment scheduling, which easily leads to unnecessary shutdowns or over-maintenance. Traditional methods rely on simple statistical models or empirical rules and do not fully utilize modern machine learning and intelligent optimization technologies.

[0006] To solve the above technical problems, the present invention provides the following technical solution: A method for intelligent operation risk analysis based on deep learning, comprising:

[0007] Collect the original operation data of power plant equipment in real time; perform data fusion on the collected data to extract equipment evaluation data; analyze the equipment evaluation data through a long short-term memory neural network model, and predict equipment failures based on the analysis results; based on the equipment failure prediction results, use the particle swarm optimization algorithm to generate an equipment scheduling plan and optimize the scheduling plan; convert the optimized equipment scheduling plan into control instructions, and generate and execute the equipment scheduling through a low-code development platform.

[0008] As a preferred solution of the intelligent operation risk analysis method based on deep learning according to the present invention, wherein: the extraction of equipment evaluation data includes performing wavelet transform denoising on the original data, outputting a pure equipment operation signal after removing high-frequency noise; optimizing the denoised signal using Kalman filtering, and outputting the equipment state data after filtering and correction; performing weighted fusion on the multi-sensor data optimized by Kalman filtering and the original data of other sensors, assigning weights based on the importance of each sensor, and outputting the fused multi-dimensional data set; using principal component analysis to reduce the dimension of the fused data set, extracting the key information representing the equipment operation characteristics, and outputting the time-domain feature data after dimension reduction; obtaining the feature information of the data in the frequency dimension through fast Fourier transform, and outputting the frequency-domain feature data; performing weighted fusion on the time-domain feature data obtained by PCA and the frequency-domain feature data obtained by FFT to generate equipment evaluation data.

[0009] As a preferred solution of the intelligent operation risk analysis method based on deep learning according to the present invention, wherein: the analysis of equipment evaluation data includes inputting the equipment evaluation data into a long short-term memory neural network model, the model includes multiple stacked LSTM layers for extracting the time-series characteristics in equipment operation; adopting a bidirectional LSTM structure so that the model can consider both the historical state and future trend of the equipment; based on the feature information output by the LSTM model, performing weighted processing through an attention mechanism to automatically focus on the key time steps of fault occurrence; using the optimized LSTM model to predict faults for the equipment evaluation data, and generating the probability of equipment failure or the prediction result of the remaining life.

[0010] As a preferred solution of the intelligent operation risk analysis method based on deep learning according to the present invention, wherein: the use of the particle swarm optimization algorithm to generate an equipment scheduling plan includes using the predicted failure probability or remaining life as the basis for evaluating the current state health risk of the equipment; generating a corresponding equipment health status score based on the equipment health risk value, and establishing a health risk assessment model; using the particle swarm optimization algorithm to optimize the equipment health risk and adjusting the scheduling strategy according to the equipment health status score; optimizing the maintenance and operation plan of the equipment through the scheduling plan generated by particle swarm optimization to maximize the equipment operation efficiency and minimize the failure risk.

[0011] As a preferred solution of the intelligent operation risk analysis method based on deep learning according to the present invention, wherein: optimizing the scheduling plan includes using the equipment health status score as the input of the particle swarm optimization algorithm to evaluate the failure probability, remaining life and health status of each device; through iterative optimization of the particle swarm algorithm, obtaining the scheduling priorities and maintenance plans of different devices; during the optimization process, considering the importance of the device, failure risk and maintenance resources as constraints to generate the optimal scheduling plan for the device; according to the optimized scheduling plan, adjusting the operation sequence, maintenance time and inspection period of the device.

[0012] As a preferred solution of the intelligent operation risk analysis method based on deep learning according to the present invention, wherein: the low-code development platform provides modular configuration of device scheduling instructions through a graphical interface.

[0013] As a preferred solution of the intelligent operation risk analysis method based on deep learning according to the present invention, wherein: generating and executing device scheduling includes, based on the optimized device scheduling plan, generating a scheduling instruction format, including device start / stop, load adjustment and maintenance scheduling; through the low-code development platform, using a graphical interface to modularly configure the scheduling instructions to form an executable operation process; converting the generated scheduling instructions into executable control instructions through the low-code development platform and sending them to relevant devices to command the devices to operate according to the predetermined scheduling plan.

[0014] Another object of the present invention is to provide an intelligent operation risk analysis system based on deep learning, which can solve the problems of inaccurate equipment failure prediction, low scheduling efficiency and slow system response by constructing an intelligent operation risk analysis system based on deep learning.

[0015] To solve the above technical problems, the present invention provides the following technical solution: an intelligent operation risk analysis system based on deep learning, including: a collection module for real-time collecting the original operation data of power plant equipment; an extraction module for performing data fusion on the collected data to extract equipment evaluation data; an analysis module for analyzing the equipment evaluation data through a long short-term memory neural network model and predicting equipment failures based on the analysis results; a scheduling module for generating an equipment scheduling plan using the particle swarm optimization algorithm based on the equipment failure prediction results and optimizing the scheduling plan; a control module for converting the optimized equipment scheduling plan into control instructions and generating and executing device scheduling through a low-code development platform.

[0016] A computer device includes a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, the steps of the intelligent operation risk analysis method based on deep learning as described above are implemented.

[0017] A computer-readable storage medium stores a computer program thereon, and when the computer program is executed by a processor, the steps of the above-described intelligent operation risk analysis method based on deep learning are implemented.

[0018] Advantages of the present invention: The intelligent operation risk analysis method based on deep learning provided by the present invention improves the fault prediction accuracy of power plant equipment and the efficiency of equipment scheduling by combining deep learning algorithms. The fault prediction based on the LSTM model can fully capture the temporal characteristics and potential fault risks in equipment operation, thereby early warning equipment faults, reducing the occurrence of sudden faults, and reducing equipment downtime and maintenance costs. Through the particle swarm optimization algorithm, the scheduling plan can be optimized according to the health status and fault risks of the equipment, improving the operation efficiency of the equipment, and effectively arranging maintenance and repair plans, avoiding the problems of over-maintenance or maintenance lag. In addition, the application of the low-code development platform simplifies the generation and execution process of the equipment scheduling plan, making the equipment scheduling more flexible and efficient. Description of the Drawings

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained according to these drawings.

[0020] Figure 1 It is the overall flowchart of an intelligent operation risk analysis method based on deep learning provided by an embodiment of the present invention. Detailed Embodiments

[0021] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will make a detailed description of the specific embodiments of the present invention in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0022] Many specific details are set forth in the following description in order to fully understand the present invention, but the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0023] Embodiment 1

[0024] Refer to Figure 1, which is an embodiment of the present invention, provides a method for analyzing intelligent operation risks based on deep learning, including:

[0025] Step S1: Real-time collect the original operation data of power plant equipment; in Step S1, the collected data includes equipment operation state parameters such as temperature, pressure, vibration, and power, and can be extended to other relevant operation state parameters according to actual needs.

[0026] Step S2: Perform data fusion on the collected data and extract equipment evaluation data; in Step S2, extracting equipment evaluation data includes performing wavelet transform denoising on the original data, removing high-frequency noise and outputting a pure equipment operation signal; using Kalman filter to optimize the denoised signal and output the equipment state data after filtering and correction; performing weighted fusion on the multi-sensor data optimized by Kalman filter and the original data of other sensors, assigning weights based on the importance of each sensor, and outputting the fused multi-dimensional data set; using principal component analysis to reduce the dimension of the fused data set, extracting the key information representing the equipment operation characteristics, and outputting the time-domain feature data after dimension reduction; obtaining the feature information of the data in the frequency dimension through fast Fourier transform and outputting the frequency-domain feature data; performing weighted fusion on the time-domain feature data obtained by PCA and the frequency-domain feature data obtained by FFT to generate equipment evaluation data.

[0027] Step S2.1: Wavelet transform denoising

[0028] Input the original equipment operation data x t , and output the denoised signal It is expressed as:

[0029]

[0030] Among them, represents performing wavelet transform on x t , λ is the level of wavelet transform, T represents the threshold function used to suppress the noise coefficient, represents the inverse wavelet transform, represents the denoised signal.

[0031] Step S2.2: Kalman filter optimization

[0032] Input the denoised signal and output the equipment operation data after Kalman filter It is expressed as:

[0033]

[0034] Among them, represents the state estimate value at the current moment k, A represents the state transition matrix, B represents the control input matrix, uk Represents the control input, K k Represents the Kalman gain, z k Represents the observed value (here it is the denoised signal ), and C represents the observation matrix.

[0035] Step S2.3: Multidimensional data weighted fusion

[0036] Input the data after Kalman filtering (Data of the i-th sensor at time k, from step S2.2) and data x of other sensors k,j (Original data of the j-th sensor at time k, without Kalman filtering), output: Multidimensional data x after weighted fusion fused,k , which is expressed as:

[0037]

[0038] where, x fused,k Represents the fused data at time k, w i Represents the weight coefficient of sensor i, satisfying Represents the data after Kalman filtering of sensor i at time k, x k,j Represents the original data of sensor j at time k.

[0039] Step S2.4: Principal component analysis (PCA) dimensionality reduction

[0040] Input the multidimensional data x after weighted fusion fused,k , apply PCA to reduce the high-dimensional data to a low-dimensional space, retain the main components, output: Device evaluation data X after dimensionality reduction PCA , as the input of step S2.5, which is expressed as:

[0041] X PCA = X fused ·V

[0042] where, X fused Represents the fused data matrix at all times k, with dimension T×D (T is the number of time steps, D is the fused feature dimension), V represents the principal component matrix of PCA, with dimension D×P (P < D, representing the number of retained principal components), X PCA Represents the device evaluation feature data after dimensionality reduction, with dimension T×P.

[0043] Step S2.5: Frequency domain feature extraction (FFT)

[0044] Input the multidimensional data x after weighted fusion fused,k , for the time-domain data x after weighted fusion fused,,kApply the Fast Fourier Transform (FFT) to convert the signal from the time domain to the frequency domain, extract the frequency domain features, and output the frequency domain feature data X FFT , which is expressed as:

[0045] X FFT = FFT(x fused,k )

[0046] where FFT(·) represents the Fast Fourier Transform operation, and X FFT represents the frequency domain feature data with a dimension of T×F (F is the frequency domain feature dimension).

[0047] Step S2.6: Multi-feature fusion and generation of final device evaluation data

[0048] Input the time domain features X PCA after dimensionality reduction and the frequency domain features X FFT . Weightedly fuse the time domain features X PCA after PCA dimensionality reduction and the frequency domain features X FFT extracted by FFT to form the final device evaluation data, and output the final device evaluation data X final , which is expressed as:

[0049] X final = α·X PCA + β·X FFT

[0050] where α and β are fusion weight coefficients satisfying α + β = 1, and X final represents the final device evaluation data with a dimension of T×(P + F).

[0051] Step S3: Analyze the device evaluation data through a long short-term memory neural network model and predict device failures based on the analysis results; in Step S3, analyze the device evaluation data, including inputting the device evaluation data into the long short-term memory neural network model, the model includes multiple stacked LSTM layers for extracting the temporal features during device operation; adopt a bidirectional LSTM structure so that the model can consider both the historical state and future trends of the device simultaneously, enhancing the temporal sensitivity to device failure prediction; based on the feature information output by the LSTM model, perform weighted processing through an attention mechanism to automatically focus on the critical time steps of fault occurrence and optimize the prediction accuracy; use the optimized LSTM model to predict device failures for the device evaluation data and generate the probability of device failure or the prediction result of the remaining life.

[0052] Step S3.1: Data preparation and preprocessing

[0053] Input the final device evaluation data X finalConvert to a format suitable for the input of the optimized LSTM model, ensuring the quality of the data and the effectiveness of model training, expressed as:

[0054]

[0055] where μ is the mean of the data and σ is the standard deviation of the data.

[0056] According to the operating cycle and characteristics of the power plant equipment, construct a time series window suitable for capturing fault precursors. Select the data of the past 60 minutes as a time window to capture short-term anomalies, expressed as:

[0057]

[0058] where t is the current time step and T = 60 is the time step length.

[0059] Output the preprocessed time series data Input as the input of the optimized LSTM model.

[0060] Step S3.2: Construction of the optimized LSTM model

[0061] Design and construct an optimized LSTM neural network model suitable for power plant equipment fault prediction. By introducing a customized attention mechanism and a multi-task learning framework, enhance the model's attention to key features and multi-dimensional prediction ability.

[0062] Construct a multi-layer LSTM model with a customized attention mechanism and multi-task learning ability, combined with a fully connected (Dense) layer and a custom activation function to form an end-to-end prediction network.

[0063] The model architecture is expressed as:

[0064]

[0065] LSTM Layer 1: h 1 = LSTM 1 (Input, dropout = 0.2, return_sequences = True)

[0066] CustomAttention Layer: h att = CustomAttention(h 1 , DomainKnowledge)

[0067] LSTM Layer 2: h 2 = LSTM 2 (h att, dropout = 0.2, return_sequences = False)

[0068] Dense Layer(FaultClassification): y class = Dense class (h 2 , activation ='sigmoid')

[0069] Dense Layer(Fault Severity): y severity = Dense severity (h 2 , activation = 'r elu ')

[0070]

[0071] Among them, LSTM 1 and LSTM 2 are the first and second LSTM units respectively, and the number of hidden units is 50 and 100 respectively; CustomAttentionLayer represents a customized attention mechanism layer, which combines the domain knowledge of power plant equipment operation (such as vibration frequency, temperature change rate), dynamically calculates the weights at each time step, and enhances the model's attention to the features of key time steps, expressed as:

[0072]

[0073] Among them, α t represents the attention weight, and DomainKnowledge t is the domain knowledge feature at time t (such as vibration frequency, temperature change rate, etc.); Concat represents the concatenation operation of features; Dense is a fully connected layer used to calculate the attention weight.

[0074] The model simultaneously performs fault category prediction (classification task) and fault severity prediction (regression task), and realizes the sharing and reuse of features by sharing the LSTM layer and the attention mechanism layer.

[0075] Introduce Dropout (such as 0.2) in the LSTM layer to reduce the risk of overfitting and improve the generalization ability of the model.

[0076] In view of the characteristics of power plant equipment operation, the following substantial algorithm adjustments and optimizations have been carried out:

[0077] Combined with the domain knowledge of equipment operation (such as vibration frequency, temperature change rate, etc.), a customized attention mechanism is designed to enable the model to dynamically adjust the attention weights according to different equipment characteristics, enhancing the ability to identify key features. The model is designed to predict both the fault category and the fault severity simultaneously, enhancing the model's feature learning ability and the comprehensiveness of prediction, enabling the model to provide more detailed fault information to assist in operation and maintenance decision-making. An adaptive learning rate adjustment strategy (such as learning rate decay based on performance metrics) is adopted to dynamically adjust the learning rate according to the model's performance on the validation set, improving the model training efficiency and stability. Through the Bayesian optimization method, the key hyperparameters of the model (such as the number of LSTM units, the number of layers, the Dropout rate, etc.) are systematically adjusted to ensure the best performance of the model in the power plant equipment fault prediction task.

[0078] Output the constructed optimized LSTM model structure, which has a customized attention mechanism and multi-task learning ability, and is specifically adapted to the requirements of power plant equipment fault prediction.

[0079] In this embodiment, the optimized long short-term memory neural network (LSTM) model consists of seven main layers, and each layer is designed and optimized for the specific requirements of power plant equipment fault prediction to improve the accuracy and practicality of prediction.

[0080] First, the input layer is responsible for receiving the preprocessed equipment evaluation time series data. This data is normalized and has a dimension of T×(P + F), where T represents the time step, and P and F represent the time domain features after PCA dimensionality reduction and the frequency domain features extracted by FFT, respectively. This layer ensures that the data received by the model has a unified scale and rich feature information, laying a solid foundation for the subsequent deep learning process.

[0081] Next, the first layer of LSTM units contains 50 hidden units and sets a Dropout rate of 0.2, and returns the sequence output of all time steps. This layer is mainly used to capture the short-term time dependencies in the input data, extract preliminary time series features, process the dynamic changes of equipment operation data, and help the model understand the equipment operation status at different time points.

[0082] The third layer is a customized attention mechanism layer, which combines the domain knowledge of power plant equipment operation, such as vibration frequency and temperature change rate, etc., by dynamically calculating the attention weight α for each time step t . This design enables the model to more accurately identify and focus on the time steps and features that are most critical for fault prediction, enhancing the model's ability to identify key abnormal patterns and improving the accuracy and timeliness of prediction.

[0083] The fourth-layer LSTM unit contains 100 hidden units, and a Dropout rate of 0.2 is also set. However, it does not return sequence outputs, but only outputs the hidden state of the last time step. This layer further captures the high-level temporal features output by the attention layer, integrates key feature information, and processes the long-term dependencies of the device operation data, enabling the model to understand the operation mode and potential fault trends of the device over a relatively long time range.

[0084] The fifth layer is a fully connected layer for fault classification, using the Sigmoid activation function, which is used to perform binary classification prediction (normal or faulty) on the device state and output the probability of a fault occurring. The sixth layer is a fully connected layer for fault severity, using the ReLU activation function, which is used to perform regression prediction on the severity of the fault (such as mild, moderate, severe), providing more detailed fault assessment information. These two layers together achieve multi-task learning, enabling the model to not only determine whether a device has a fault but also evaluate the severity of the fault, providing a more comprehensive reference basis for operation and maintenance decisions.

[0085] Finally, the output layer outputs the fault category prediction result and the fault severity prediction result respectively. These output results will be used in the real-time fault warning system to notify the operation and maintenance personnel to take corresponding measures in a timely manner to ensure the safe and stable operation of the device.

[0086] Based on the traditional LSTM model, the present invention significantly improves the performance and practicality of power plant equipment fault prediction through the following specific improvements. First, the customized attention mechanism combines the domain knowledge of power plant equipment operation, enabling the model to dynamically adjust the attention weights according to different device characteristics, accurately capture the key time steps and features, significantly enhancing the ability to identify key abnormal patterns during device operation, and improving the accuracy and timeliness of fault prediction. Second, the multi-task learning framework enables the model to simultaneously perform fault category prediction and fault severity prediction. By sharing feature representations, it enhances the learning efficiency and feature expression ability of the model, provides more comprehensive fault information, and aids in more effective operation and maintenance decisions.

[0087] In addition, the combination of adaptive learning rate adjustment and early stopping mechanism optimizes the model training process, improves training efficiency, prevents overfitting, and ensures good performance of the model under different data distributions. Systematic hyperparameter optimization automatically adjusts the key hyperparameters of the model through Bayesian optimization method, further enhancing the prediction performance and stability of the model, and ensuring the adaptability and reliability of the model under different devices and operating environments. Finally, multi-level data processing and feature fusion generate high-quality and low-noise device evaluation data through wavelet transform denoising, Kalman filter optimization, weighted data fusion, PCA dimensionality reduction, and FFT frequency domain feature extraction, ensuring that the data input to the LSTM model is accurate and information-rich, enhancing the prediction performance and fault identification ability of the model, and reducing the impact of data noise and interference on the prediction results.

[0088] Step S3.3: Model Training and Optimization

[0089] Use historical device operation data to train the optimized LSTM model so that it can learn the fault patterns and time dependencies in device operation.

[0090] Divide the preprocessed data into a training set, a validation set, and a test set in a ratio of 70% for training, 15% for validation, and 15% for testing to ensure the effectiveness of model training and the reliability of evaluation. Select 32 as the batch size to balance training speed and memory consumption. Set 100 epochs of training and combine it with the early stopping mechanism to decide whether to stop early according to the performance of the validation set. Monitor the loss of the validation set. When the validation loss does not decrease for 10 consecutive epochs, stop training early to prevent overfitting.

[0091] Model optimization includes automatically adjusting hyperparameters such as the learning rate, regularization parameter, and number of LSTM units through the Bayesian optimization method to find the optimal combination. Introduce a Dropout layer (such as 0.2) in the LSTM layer to reduce the risk of model overfitting. For the problem of class imbalance in power plant equipment data, adjust the weights of different tasks in the loss function to enhance the model's ability to identify minority classes (faults), expressed as:

[0092]

[0093] where α and β are adjusted according to the prediction requirements of the importance and severity of the fault categories to ensure a reasonable ratio of losses for each task.

[0094] Output: An optimized LSTM model with completed training and optimal performance, having high-accuracy and stable fault prediction capabilities.

[0095] Step S3.4: Model Evaluation and Validation

[0096] Evaluate the performance of the trained optimized LSTM model on the test set to ensure its good generalization ability and fault prediction accuracy.

[0097] The performance metric evaluation includes:

[0098] Accuracy:

[0099]

[0100] Precision:

[0101]

[0102] Recall:

[0103]

[0104] F1-Score:

[0105]

[0106] Mean Squared Error (MSE) (for fault severity prediction):

[0107]

[0108] ROC curve and AUC value are used to evaluate the classification ability of the model at different thresholds.

[0109] Confusion matrix analyzes the performance of the model in fault class prediction, identifies false positive and false negative patterns, and guides subsequent model optimization.

[0110] Plot the loss curve and accuracy curve of the training and validation sets to observe the convergence situation and whether there is overfitting during the model training process.

[0111] Plot the ROC curve, analyze the classification performance of the model, calculate the AUC value, and evaluate the overall classification ability of the model.

[0112] Compare the optimized LSTM model with traditional LSTM models and other deep learning models (such as GRU, CNN-LSTM) to verify its superiority in fault prediction tasks.

[0113] Output the model evaluation report, including various performance metrics and visualization charts, to confirm the prediction ability and stability of the optimized model.

[0114] Step S3.5: Real-time fault prediction and warning

[0115] Use the trained optimized LSTM model to predict faults in real-time device operation data, issue early warnings, and ensure the safe and stable operation of the device.

[0116] Perform the same preprocessing steps on the real-time collected device operation data as the training set, including wavelet transform denoising, Kalman filter optimization, data fusion, PCA dimensionality reduction, FFT frequency domain feature extraction, and multi-feature fusion, to generate

[0117] The preprocessed real-time data Input it into the optimized LSTM model to obtain the fault prediction result, expressed as:

[0118]

[0119] Among them, is the real-time prediction result, including the fault category (normal / fault) and the fault severity (mild / moderate / severe).

[0120] According to the prediction result, if a fault is predicted, trigger the warning system to notify the operation and maintenance personnel to take corresponding measures. The warning mechanism includes: SMS notification, email reminder, and system interface alarm to ensure that the operation and maintenance personnel can receive the warning information in a timely manner. The fault prediction result and the warning status are displayed in real-time on the monitoring platform for the operation and maintenance personnel to respond quickly. Regularly collect new device operation data and fault records, retrain and optimize the LSTM model to ensure the continuous improvement of its prediction ability and adaptability. Adopt the incremental learning method to gradually update the model parameters using new data, reducing the time and resource consumption of retraining. Collect the feedback from the operation and maintenance personnel on the warning results, adjust and optimize the model to improve the accuracy and practicality of the prediction.

[0121] Specifically, aiming at the multi-source heterogeneity and time-series complexity of power plant device operation data, the method of this embodiment makes customized algorithm adjustments based on the traditional LSTM model. First, introduce an attention mechanism enhanced by domain knowledge, and input the unique operation parameters of power plant devices (such as vibration frequency, temperature change rate, etc.) as auxiliary features into the attention layer, enabling the model to dynamically adjust the attention weights for key time steps and key features. This customized attention mechanism not only improves the model's ability to identify device abnormal patterns but also enhances the capture accuracy of potential fault precursors. Second, adopt a multi-task learning framework, where the model simultaneously predicts the fault category (normal / fault) and the fault severity (mild / moderate / severe). By sharing the LSTM layer and the attention mechanism layer, feature sharing and reuse are realized, improving the model's learning efficiency and feature expression ability.

[0122] To further improve the training efficiency and generalization ability of the model, the method of this embodiment introduces an adaptive learning rate adjustment strategy and an early stopping mechanism. Specifically, a learning rate decay method based on the performance of the validation set is adopted. When the loss of the model on the validation set no longer decreases for several consecutive rounds, the learning rate is automatically reduced to accelerate the convergence speed of the model. At the same time, combined with the early stopping mechanism, when the validation set loss does not improve within the set patience parameter, the training is terminated in advance to prevent the model from overfitting. In addition, the key hyperparameters of the model (such as the number of LSTM units, the number of layers, the Dropout rate, etc.) are systematically adjusted and optimized through the Bayesian optimization method to ensure the best performance of the model in the power plant equipment fault prediction task. These adaptive training strategies and systematic hyperparameter optimizations significantly improve the stability and prediction accuracy of the model.

[0123] It should be noted that the method of this embodiment combines the specific requirements and characteristics of the operation of power plant equipment and solves the specific technical problems brought by the high-dimensionality and time-series nature of multi-source heterogeneous data. Through a customized attention mechanism and a multi-task learning framework, the model can more accurately identify the key abnormal patterns in the equipment operation and achieve the simultaneous prediction of the fault category and severity. This optimization process ensures the high quality and rich information of the input data and improves the prediction performance and robustness of the model.

[0124] Step S4: Based on the equipment fault prediction results, use the particle swarm optimization algorithm to generate an equipment scheduling plan and optimize the scheduling plan; in step S4, using the particle swarm optimization algorithm to generate an equipment scheduling plan includes using the predicted fault probability or remaining life as the basis for evaluating the current state health risk of the equipment; generating a corresponding equipment health status score based on the equipment health risk value and establishing a health risk assessment model; using the particle swarm optimization algorithm to optimize the equipment health risk and adjusting the scheduling strategy according to the equipment health status score; optimizing the maintenance and operation plans of the equipment through the scheduling plan generated by the particle swarm optimization to maximize the equipment operation efficiency and minimize the fault risk.

[0125] Optimizing the scheduling plan includes using the equipment health status score as the input of the particle swarm optimization algorithm to evaluate the fault probability, remaining life and health status of each equipment; obtaining the scheduling priorities and maintenance plans of different equipment through iterative optimization of the particle swarm algorithm; considering the importance of the equipment, the fault risk and the maintenance resources as constraints during the optimization process to generate the optimal scheduling plan for the equipment; adjusting the operation sequence, maintenance time and overhaul cycle of the equipment according to the optimized scheduling plan.

[0126] The health status score comprehensively considers the fault probability and remaining life of the equipment and is expressed as:

[0127]

[0128] Among them, H score (i) represents the health status score of device i, and P fault (i) represents the failure probability of device i, with a value range of [0, 1], and R life (i) represents the remaining life of device i, and R life,max represents the maximum design life of the device, serving as a normalization reference for the remaining life. α and β represent weight coefficients used to balance the importance of the failure probability and the remaining life in the health score, and α + β = 1.

[0129] The health risk assessment model is expressed as:

[0130] R health (i) = γ·H score (i) + δ

[0131] Among them, R health (i) represents the health risk value of device i, and γ and δ represent constants used to adjust the sensitivity and baseline value of the health risk assessment model.

[0132] Use the particle swarm optimization algorithm to optimize the device health risk and adjust the scheduling strategy according to the device health status score. The specific steps are as follows:

[0133] I. PSO parameter initialization:

[0134] Set the basic parameters of the PSO algorithm: Particle swarm size (N): such as 50. Maximum number of iterations (T): such as 100. Inertia weight (w): such as 0.5. Individual learning factor (c 1 ) : such as 1.5. Global learning factor (c 2 ) : such as 1.5.

[0135] II. Particle encoding

[0136] Each particle represents a device scheduling scheme and is encoded as a vector:

[0137] X i = {x i1 , x i2 , …, x in}

[0138] Among them, X i represents the scheduling scheme of particle i, and x ij represents the scheduling parameter of device j, such as start time, stop time, maintenance time, etc.

[0139] III. Definition of the objective function

[0140] Define the objective function of PSO to maximize the device operation efficiency and minimize the failure risk:

[0141] f(X i ) = w 1 ·Efficiency(X i ) - w 2 ·Risk(X i )

[0142]

[0143] Wherein, Efficiency(X i ) represents the equipment operation efficiency under the scheduling scheme X i , Risk(X i ) represents the total failure risk under the scheduling scheme X i , w 1 and w 2 represent the weight coefficients, respectively indicating the importance of operation efficiency and failure risk in the objective function, O j represents the effective operation time of equipment j, and T j represents the total operation time of equipment j.

[0144] IV. PSO Iterative Optimization

[0145] Iteratively optimize the positions and velocities of the particles through the following steps to find the optimal scheduling scheme:

[0146] Initialize the particle swarm: Randomly generate the initial position X i 0 and velocity v i 0 .

[0147] Evaluate the fitness and calculate the fitness value f(X i ) of each particle according to the objective function.

[0148] Update the individual best position and the global best position; for each particle i, if the current fitness value is better than the historical best fitness value, then update the individual best position p best,i ; Update the global best position g best of the swarm, that is, the position of the particle with the highest fitness value in the swarm.

[0149] Update the velocity and position, and update the velocity and position of the particle according to the following formula:

[0150]

[0151] Wherein, and are respectively the velocity and position of particle i in the k-th round, r 1 and r 2 are random numbers in the range of [0, 1], pbest,i and g best are the historical best position of particle i and the global best position of the population, respectively.

[0152] Constraint handling: After each position update, ensure that the scheduling plan meets the preset constraints:

[0153] Importance(j) ≥ γ, Risk(j) ≤ δ, Resource(j) ≤ θ

[0154] where γ represents the importance threshold of device j, δ represents the risk threshold of device j, and θ represents the maintenance resource threshold of device j.

[0155] Stop the iteration when any of the following conditions is met:

[0156] The maximum number of iterations T is reached or the global best fitness value has not improved significantly in consecutive rounds of iterations (set a tolerance threshold).

[0157] V. Generation of the Optimal Scheduling Plan

[0158] Through the iterative process of the particle swarm optimization algorithm, the optimal device scheduling plan is finally obtained:

[0159] X optimal ={x 1,optimal ,x 2,optimal ,…,x n,optimal}

[0160] where x j,optimal includes scheduling parameters such as the operating sequence, repair time, and maintenance cycle of device j:

[0161] Schedule(j)={StartTime(j), StopTime(j), RepairTime(j), MaintenanceCycle(j)}

[0162] Combining the device failure probability and remaining life predicted by the long short-term memory neural network model with the particle swarm optimization algorithm, a comprehensive health status scoring and health risk assessment model is formed. It can quantify the health status of each device, organically integrating the two key indicators of failure probability and remaining life, improving the accuracy of device health assessment. By establishing a health risk assessment model, this method can further refine the health risk level of the device, ensuring that during the particle swarm optimization process, the scheduling scheme can fully consider the actual operating status and potential risks of the device. Under multi-dimensional constraint conditions, the PSO algorithm uses the health risk score as input and generates the optimal device scheduling scheme through iterative search. This process not only maximizes the device operation efficiency but also effectively minimizes the failure risk, significantly enhancing the intelligent and optimized level of power plant device management and ensuring the stability and reliability of power plant operation.

[0163] Furthermore, during the optimization process of the scheduling scheme, this method comprehensively considers multiple constraint conditions such as the importance of the device, failure risk, and maintenance resources, ensuring that the generated scheduling scheme is both efficient and meets the actual operation requirements. Through the iterative optimization of the particle swarm optimization algorithm, the system can dynamically adjust the operation sequence, maintenance time, and inspection cycle of the device to achieve the best balance between device maintenance and operation plans. In addition, the introduction of the health risk assessment model enables the scheduling scheme to have stronger adaptability and response capabilities when dealing with sudden failures and resource limitations.

[0164] Step S5: Convert the optimized device scheduling scheme into control instructions and generate and execute device scheduling through a low-code development platform.

[0165] In step S5, the low-code development platform provides modular configuration of device scheduling instructions through a graphical interface.

[0166] The low-code development platform includes: a device scheduling rule library for storing preset rules and scheduling templates related to device scheduling; a scheduling instruction automatic generation module for generating corresponding scheduling control instructions based on the device failure prediction results and the scheduling rule library; and a user interaction interface for customizing scheduling templates and adjusting instruction parameters according to user needs to optimize the scheduling strategy.

[0167] Based on the optimized device scheduling scheme, generate a scheduling instruction format, including device start / stop, load adjustment, and maintenance scheduling; through the low-code development platform, use the graphical interface to modularly configure the scheduling instructions to form an executable operation process; convert the generated scheduling instructions into executable control instructions through the low-code development platform and send them to the relevant devices to command the devices to operate according to the predetermined scheduling scheme.

[0168] Embodiment 2

[0169] An embodiment of the present invention provides a smart operation risk analysis system based on deep learning, including:

[0170] An acquisition module for real-time acquisition of the original operation data of power plant equipment;

[0171] An extraction module for data fusion of the acquired data and extraction of equipment evaluation data;

[0172] An analysis module for analyzing the equipment evaluation data through a long short-term memory neural network model and predicting equipment failures based on the analysis results;

[0173] A scheduling module for generating an equipment scheduling plan using a particle swarm optimization algorithm based on the equipment failure prediction result and optimizing the scheduling plan;

[0174] A regulation module for converting the optimized equipment scheduling plan into a control instruction and generating and executing equipment scheduling through a low-code development platform.

[0175] Embodiment 3

[0176] An embodiment of the present invention, which is different from the previous two embodiments, is as follows:

[0177] If the said function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that makes a contribution to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. And the aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0178] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a definitional sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device), or used in combination with these instruction execution systems, apparatuses, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.

[0179] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection portion having one or more wirings (electronic device), a portable computer disk cartridge (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, a computer-readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpretation, or other suitable processing as necessary, and then stored in a computer memory.

[0180] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application-specific integrated circuits having suitable combinational logic gate circuits, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), and the like.

[0181] Embodiment 4

[0182] For an embodiment of the present invention, a deep learning-based intelligent operation risk analysis method is provided. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through simulation experiments.

[0183] In this embodiment, the operating environment of power plant equipment is set, and the operating data of the equipment is collected in this environment, including vibration, temperature, pressure, and power parameters. The same set of power plant equipment is used as the experimental object in the experiment to ensure the comparability of the experimental results. The equipment used in the experiment has been comprehensively fault-detected before the experiment to ensure that it has a certain fault risk throughout the experimental period and can reflect the effects of fault prediction and scheduling optimization. To ensure the scientificity and effectiveness of the experiment, the data collection period is 30 days, and the operating state of the equipment is continuously monitored during the experiment. The number of experimental equipment is 50, covering different types of power generation equipment, and the data collection frequency is once per minute. The main objectives of the experiment are to evaluate the fault prediction accuracy, scheduling efficiency, maintenance cost, and operating efficiency.

[0184] During the experiment of the traditional method, first, the operating data of the equipment is real-time monitored through a standard sensor data collection system. After the collected raw data is simply cleaned, fault prediction is performed through traditional statistical analysis methods. These statistical methods mainly include linear regression analysis and rule-based fault detection algorithms, and the model judges whether the equipment has a fault based on historical data and thresholds. On the basis of prediction, the traditional method uses a rule-based scheduling algorithm, combined with the fault prediction results of the equipment and the existing maintenance strategies to perform scheduling. The scheduling process of this method is as follows: manually adjust the operation and maintenance order of the equipment according to the equipment fault prediction results, and perform scheduling based on the importance and availability of the equipment. The scheduling instructions are written manually and sent to the equipment through a traditional control system to perform corresponding operations. The key steps in this process include fault prediction, equipment health status assessment, scheduling plan generation, and scheduling execution. There is a large amount of manual intervention in the whole process, and manual judgment and operation are required according to the equipment fault prediction results. The scheduling response time is long, and it is difficult to adapt to the dynamically changing equipment state.

[0185] The experimental process of the method of the present invention is optimized compared with the traditional method. First, after the original data of the device is collected in real time, it is fused through a variety of data preprocessing techniques, including wavelet transform for noise reduction, Kalman filter for optimizing the noise-reduced signal, principal component analysis (PCA) for dimensionality reduction of the data, and fast Fourier transform (FFT) for extracting the frequency-domain characteristics of the device operation signal. After these processes, the device evaluation data is input into the long short-term memory neural network (LSTM) model. The model extracts the temporal characteristics during the device operation through multiple stacked LSTM layers and combines the bidirectional LSTM structure to enhance the perception of historical and future states. The LSTM model automatically focuses on the critical time steps of device failures through the attention mechanism and generates the probability of device failures and the prediction results of the remaining life. After obtaining the failure prediction results, the method of the present invention uses the particle swarm optimization algorithm (PSO) to generate and optimize the scheduling scheme of the device. The PSO algorithm generates the optimal scheduling scheme according to the health risk assessment value of the device, the failure prediction results, and the maintenance requirements of the device. After the scheduling scheme is generated, with the help of a low-code development platform, the device scheduling instructions are quickly generated and modularly configured through a graphical interface, and finally the instructions are sent to the device for execution. The optimization of the scheduling scheme not only considers the health risk of the device, but also optimizes the maintenance cycle and operation sequence of the device to minimize the failure risk and improve the operation efficiency of the device. The scheduling response time and execution efficiency are greatly improved compared with the traditional method, and the human intervention is reduced. The experimental results are shown in Table 1.

[0186] Table 1 Comparison Table of Experimental Results

[0187]

[0188] The traditional method is based on statistical analysis and rule judgment, relying on the threshold judgment and linear regression of historical data, resulting in poor performance in multi-dimensional data processing. Especially when dealing with the operation data of complex devices, the traditional method is difficult to capture the potential changes in the device state, the prediction accuracy is limited, and it is difficult to cope with dynamic and complex working conditions.

[0189] The method of the present invention uses the long short-term memory network in deep learning, which can effectively mine the temporal characteristics in the device data. By using the bidirectional LSTM structure, the model considers the historical and future device states at the same time, and combines the attention mechanism, enabling the model to automatically identify and focus on the critical moments of failure occurrence in a large amount of data. These technological innovations give the method of the present invention significant advantages in failure prediction, enabling more accurate prediction of device failures and avoiding the errors and omissions of the traditional method.

[0190] In terms of equipment scheduling, traditional methods rely on manual operations and rule algorithms. The scheduling process is not efficient enough and is vulnerable to human interference, resulting in slow response speeds and an inability to quickly adapt to sudden changes during equipment operation. In contrast, the method of the present invention uses the particle swarm optimization algorithm to generate an optimal scheduling plan based on the equipment health risk score and the fault prediction results. The optimized scheduling plan improves the response speed, reduces equipment downtime, and achieves more precise optimization in equipment resource allocation, thus enhancing the overall equipment operation efficiency.

[0191] In addition, during the equipment maintenance process of traditional methods, due to the lack of scientific scheduling and optimization strategies, there is often a waste of maintenance resources and excessive consumption of equipment. The particle swarm optimization algorithm of the present invention can, while dynamically adjusting the maintenance and operation strategies of equipment, avoid over-maintenance or over-operation, saving a large amount of maintenance costs and extending the service life of the equipment.

[0192] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A smart operation risk analysis method based on deep learning, characterized in that: include: Collect the original operation data of power plant equipment in real time; Perform data fusion on the collected data and extract equipment evaluation data; Analyze equipment assessment data through long short-term memory neural network models and predict equipment failures based on the analysis results; Based on the equipment failure prediction results, the particle swarm optimization algorithm is used to generate equipment scheduling plans and optimize the scheduling plans; Convert the optimized equipment scheduling plan into control instructions, and generate and execute equipment scheduling through the low-code development platform.

2. The method for intelligent operation risk analysis based on deep learning according to claim 1, characterized in that: The extracting device evaluation data comprises, Perform wavelet transform on the original data to reduce noise, remove high-frequency noise and output pure equipment operation signal; Kalman filtering is used to optimize the noise reduction signal and output the filtered and corrected equipment status data; The multi-sensor data optimized by Kalman filtering is weightedly fused with the original data of other sensors, and weights are assigned based on the importance of each sensor, and the fused multi-dimensional data set is output; Use principal component analysis to reduce the dimension of the fused data set, extract key information representing the operating characteristics of the equipment, and output the time domain feature data after dimension reduction; Obtain the characteristic information of the data in the frequency dimension through fast Fourier transform and output the frequency domain characteristic data; The time domain feature data obtained by PCA and the frequency domain feature data obtained by FFT are weighted and fused to generate equipment evaluation data.

3. The method for intelligent operation risk analysis based on deep learning as claimed in claim 2, characterized in that: The analysis device evaluates data including, Inputting the equipment evaluation data into a long short-term memory neural network model, wherein the model includes a plurality of stacked LSTM layers for extracting timing features in equipment operation; The bidirectional LSTM structure enables the model to consider both the historical status and future trends of the equipment. Based on the feature information output by the LSTM model, the key time step of the fault occurrence is automatically focused on through weighted processing by the attention mechanism; Use the optimized LSTM model to predict failures of equipment evaluation data and generate equipment failure probability or remaining life prediction results.

4. The method for intelligent operation risk analysis based on deep learning according to claim 3, characterized in that: The method of using a particle swarm optimization algorithm to generate a device scheduling solution includes: Use the predicted probability of failure or remaining life as a basis for assessing the health risk of the equipment’s current state; Based on the equipment health risk value, generate the corresponding equipment health status score and establish a health risk assessment model; Use particle swarm optimization algorithm to optimize equipment health risks and adjust scheduling strategies based on equipment health status scores; The scheduling scheme generated by particle swarm optimization is used to optimize the maintenance and operation plan of the equipment to maximize the equipment operation efficiency and minimize the risk of failure.

5. The method for intelligent operation risk analysis based on deep learning according to claim 4, characterized in that: The scheduling scheme optimization includes: The equipment health status score is used as the input of the particle swarm optimization algorithm to evaluate the failure probability, remaining life and health status of each equipment; Through iterative optimization of particle swarm algorithm, the scheduling priority and maintenance plan of different equipment are obtained; During the optimization process, the importance of the equipment, the risk of failure, and the maintenance resources are considered as constraints to generate the optimal scheduling plan for the equipment; According to the optimized scheduling plan, adjust the equipment's operating sequence, maintenance time and overhaul cycle.

6. The method for intelligent operation risk analysis based on deep learning according to claim 5, characterized in that: The low-code development platform provides modular configuration of device scheduling instructions through a graphical interface.

7. The method for intelligent operation risk analysis based on deep learning according to claim 6, characterized in that: The generating and executing device scheduling includes: Generate scheduling instruction format based on the optimized equipment scheduling plan, including equipment start and stop, load adjustment and maintenance scheduling; Through the low-code development platform, the scheduling instructions are modularly configured using a graphical interface to form an executable operation process; The generated scheduling instructions are converted into executable control instructions through the low-code development platform and sent to relevant equipment to command the equipment to operate according to the predetermined scheduling plan.

8. A system using the deep learning-based intelligent operation risk analysis method as described in any one of claims 1 to 7, characterized in that: include: The acquisition module is used to collect the original operation data of power plant equipment in real time; An extraction module is used to fuse the collected data and extract equipment evaluation data; An analysis module for analyzing equipment evaluation data through a long short-term memory neural network model and predicting equipment failures based on the analysis results; The scheduling module is used to generate equipment scheduling plans based on equipment failure prediction results using the particle swarm optimization algorithm and optimize the scheduling plans; The control module is used to convert the optimized equipment scheduling plan into control instructions, generate and execute equipment scheduling through the low-code development platform.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the deep learning-based intelligent operation risk analysis method described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the deep learning-based intelligent operation risk analysis method described in any one of claims 1 to 7 are implemented.

Images