A robust fire warning system based on multi-source time series data fusion

Abstract

The present application belongs to the technical field of fire warning, the traditional fire warning mainly uses sensor to carry out state detection and identification, although sensor technology has made remarkable progress in fire warning, but there are still many challenges in further shortening the fire warning time, reducing the false alarm rate and the false negative rate and the like, in view of these problems, the present application proposes a kind of robust fire warning system based on multi-source time series data fusion, mainly rely on three types of sensors on hardware: temperature sensor, CO concentration sensor and smoke concentration sensor, the model uses the time series data collected by temperature, CO concentration, smoke concentration sensor, extracts 12-dimensional parameters as the input of neural network for multi-source data fusion, and considers the abnormal situation of data acquisition simultaneously, compared with other existing methods, the present application greatly reduces the false alarm rate and the false negative rate of fire warning, significantly shortens the time of fire warning.

Description

Technical Field

[0001] This invention relates to the field of fire early warning technology, and in particular to a robust fire early warning system based on multi-source time-series data fusion. Background Technology

[0002] In recent years, frequent fires have occurred nationwide, causing enormous casualties and property losses. The immense harm caused by fires necessitates proactive fire prevention measures. Currently, mainstream fire detection methods include vision-based and sensor-based approaches. Compared to visual methods for fire identification, sensor-based fire detection is more applicable to most environments, offering advantages such as wide applicability, low cost, simple maintenance, and rapid response. Currently, traditional fire sensors based on temperature, smoke, and gas concentration are widely used for fire detection in indoor spaces in my country, with smoke sensors accounting for the majority. However, these sensors can only detect single environmental factors and are easily affected by interference from other factors. To reduce the interference of environmental factors on sensor information and to provide more timely and accurate alarms, we propose a robust fire early warning system based on multi-source time-series data fusion.

[0003] Currently, mainstream smoke sensors on the market determine fires by monitoring smoke concentration in a single area at a specific time. To improve the performance of fire alarm systems, data such as temperature, CO concentration, and smoke concentration are collected as input data and trained using a backpropagation neural network to reduce errors caused by interference from single environmental factors. Although this method significantly improves alarm accuracy compared to single-sensor methods, existing methods do not consider the temporal evolution of sensor data and potential acquisition anomalies. Abnormal actual values ​​can lead to significant deviations in fire warnings, resulting in a still high false alarm and false negative rate for the alarm system. Therefore, there is an urgent need to develop a highly accurate and robust fire warning system to address the shortcomings of existing research. Summary of the Invention

[0004] Given that existing fire alarm systems suffer from high false alarm rates, low accuracy, and poor robustness, this invention proposes a robust fire early warning system based on multi-source time-series data fusion.

[0005] This invention proposes a robust fire early warning system based on multi-source time-series data fusion, including...

[0006] Data anomaly handling module:

[0007] The fire early warning system includes three types of sensors in its hardware: temperature sensor, CO concentration sensor, and smoke concentration sensor. The time-series data collected at a specific point in time is used... It means that among them Represents the type of sensor. Represents a time series;

[0008] First, anomaly detection is performed on the collected time-series data to determine... Is it within the confidence interval?

[0009]

[0010] These are interval values ​​set according to actual conditions. If it exceeds this range, it means The data contains anomalies and requires anomaly handling. Experiments with missing data have shown that autoregressive models are effective for predicting unknown data. Therefore, the autoregressive formula is used to predict outlier values:

[0011]

[0012] In the autoregressive formula, This is a random error value. The autocorrelation coefficient represents the relationship between the current observation and past data. The coefficients of the linear relationship between the observations at each time point are used to predict the data at the anomalous time points, and then the predicted values ​​are... The above is the abnormal data processing module of the entire model, replacing the original abnormal data.

[0013] Neural Network Module:

[0014] Time series data of factors that show significant changes in the fire process are obtained from known datasets, including temperature, carbon monoxide concentration and smoke concentration, and feature values ​​that can accurately reflect the data change process are extracted from them.

[0015] The input data is ,express Sensor data in The value of the moment. Indicates the number of different types of sensors. The dataset consists of time-labeled sequences;

[0016] Information extraction is performed, including: moving average, baseline error, forecast baseline error, and cyclic trend indicator. The moving average is a well-known technical indicator used to predict future data in time series analysis. It enhances the judgment of data changes. The formula for the moving average is:

[0017]

[0018] express Sensor data in Moving average over time;

[0019] The moving average is a key element in extracting changes in a time series. Based on this, the baseline error and the baseline error in the prediction stage can be calculated. The baseline error is calculated as the difference between the current sensor value and the environmental baseline value. This directly reflects the degree of deviation between the current data and the environmental baseline. The environmental baseline value is represented by the moving average. The formula for the moving baseline error is:

[0020]

[0021] express Sensor data in At any time and The difference is used as an input to the neural network.

[0022] To better predict fire occurrences in advance, the model determines the error by comparing the predicted data for the next stage with a set baseline. The baseline error is then calculated using the exponential smoothing formula:

[0023]

[0024] in and The two constants are not directly related; they are used to weight historical and current values. express Sensor data in The model then extracts the cyclical trend of the sensor's previous time-based reference error, current time-based reference error, and next time-based prediction reference error during the time-prediction phase. Returns only the characteristics of -1, 0, and 1, and returns the cyclic trend indicators for different sensors:

[0025]

[0026] The cyclic trend indicator is the baseline error of the previous time step, the baseline error of the current time step, and the baseline error of the next time step. The summation operation under a function is specifically represented as follows:

[0027]

[0028] express Sensor data in The cyclical trend indicator at any given moment is a comprehensive reflection of the degree of trend change of the same sensor at the previous moment, the current moment, and the next moment;

[0029] Since the model employs a backpropagation (BP) neural network, which uses a sigmoid function as the activation function, to prevent smaller values ​​in the output data from being swallowed up by larger values, the outlier-handled data needs to be processed. and the information extracted , and Normalization is performed, and the specific normalization formula is as follows:

[0030]

[0031] in For the input vector, This represents the normalized value of the parameter input. This represents the minimum value of the input vector. This represents the maximum value of the input vector;

[0032] The model will , , , Normalized data is used as input to a neural network, and the fire probability y is used as the output. Addressing the problems of low accuracy, high false alarm rate, and poor robustness in traditional fire early warning methods, this invention proposes a robust fire early warning system based on multi-source time-series data fusion. This model utilizes the temporal characteristics of sensor data to train a BP neural network, resulting in a model with high accuracy, low false alarm rate, and low missed alarm rate. Simultaneously, it estimates and replaces abnormal data collected by sensors, effectively improving the robustness of fire early warning. By fully utilizing the temporal evolution patterns of multi-sensor data and preprocessing methods for abnormal data collection, a fire early warning model with high robustness, high accuracy, and low false alarm rate is developed.

[0033] Preferably, the model treats the input of the neural network as: , respectively represent the first The inclusion of temperature, CO concentration, smoke concentration, temperature baseline error, CO concentration baseline error, smoke concentration baseline error, temperature prediction baseline error, CO concentration prediction baseline error, smoke concentration prediction baseline error, temperature cycle trend index, CO concentration cycle trend index, and smoke concentration cycle trend index is beneficial for model formation.

[0034] Preferably, the algorithm flow of the robust fire early warning system based on multi-source time-series data fusion includes initializing the network, using the input data as the input vector of the BP neural network linear model, with each set of input data, i.e., each row, corresponding to a fire probability label, and determining the network structure as a three-layer structure: input layer, hidden layer, and output layer, to facilitate subsequent operations.

[0035] Preferably, the following is the information for each layer:

[0036] Expected output Indicates the first The expected output of the group is the fire probability;

[0037] The input to the hidden layer is: ;

[0038] The output of the hidden layer is: ;

[0039] The input to the output layer is: ;

[0040] The output of the output layer is: ;

[0041] The weights of the input layer and the hidden layer are The threshold is ;

[0042] Weights of hidden layers and output layers The threshold is ;

[0043] Randomly select a set of samples as input and target samples: the first Group input and the corresponding expected output Select error function Set the calculation precision and the number of learning iterations.

[0044] Preferred forward propagation: Input data is passed through the network, the output of each neuron is calculated, and the weighted input and the output of the activation function are calculated in each layer;

[0045] Calculate the input of each unit in the hidden layer:

[0046]

[0047] Output of the hidden layer:

[0048]

[0049] Inputs to each neuron in the output layer:

[0050]

[0051] Outputs of each neuron in the output layer:

[0052]

[0053] Preferably, the error is calculated by comparing the output of the neural network with the actual target and using an error function.

[0054] Calculate the unit error between the actual output layer and the target output layer:

[0055]

[0056] Calculate the error of each unit in the intermediate layer:

[0057] .

[0058] Preferably, backpropagation: starting from the output layer, calculate the error gradient for each layer:

[0059] The chain rule is used to calculate the contribution of each parameter to the error, and the error gradient is propagated backward to each layer to update each parameter to minimize the error:

[0060] Adjust the weights and thresholds from the output layer to the hidden layer:

[0061]

[0062]

[0063] Correct the weights and thresholds from the hidden layer to the input layer:

[0064]

[0065]

[0066] in, For learning rate, .

[0067] Preferably, the update iteration is performed by determining the number of training iterations based on the loss of the training data, and repeatedly executing forward propagation and backward propagation until the number of learning iterations reaches a preset value.

[0068] Preferably, the model is trained and simulated using Python, with the National Institute of Standards and Technology (NIST) public dataset selected as the training sample, the learning rate set to 0.001, the training batch size set to 32, and the iterations repeated 1000 times to obtain the final model.

[0069] The beneficial effects of this invention are as follows: Addressing the problems of low accuracy, high false alarm rate, and poor robustness faced by traditional fire early warning methods, this invention proposes a robust fire early warning system based on multi-source time-series data fusion. This model utilizes the temporal characteristics of sensor data to train a model with high accuracy, low false alarm rate, and low missed alarm rate through a BP neural network. Simultaneously, it estimates and replaces abnormal data collected by sensors, effectively improving the robustness of fire early warning. By fully utilizing the temporal evolution laws of multi-sensor data and preprocessing methods for abnormal data collection, a fire early warning model with high robustness, high accuracy, and low false alarm rate is developed. Attached Figure Description

[0070] Figure 1 This is a flowchart of a robust fire early warning system based on multi-source time-series data fusion proposed in this invention.

[0071] Figure 2 This is a flowchart of the algorithm for a robust fire early warning system based on multi-source time-series data fusion proposed in this invention. Detailed Implementation

[0072] The present invention will be further explained below with reference to specific embodiments.

[0073] Example

[0074] refer to Figure 1-2 This embodiment proposes a robust fire early warning system based on multi-source time-series data fusion, including...

[0075] Data anomaly handling module:

[0076] The fire early warning system includes three types of sensors in its hardware: temperature sensor, CO concentration sensor, and smoke concentration sensor. The time-series data collected at a specific point in time is used... It means that among them Represents the type of sensor. Represents a time series;

[0077] First, anomaly detection is performed on the collected time-series data to determine... Is it within the confidence interval?

[0078]

[0079] These are interval values ​​set according to actual conditions. If it exceeds this range, it means The data contains anomalies and requires anomaly handling. Experiments with missing data have shown that autoregressive models are effective for predicting unknown data. Therefore, the autoregressive formula is used to predict outlier values:

[0080]

[0081] In the autoregressive formula, This is a random error value. The autocorrelation coefficient represents the relationship between the current observation and past data. The coefficients of the linear relationship between the observations at each time point are used to predict the data at the anomalous time points, and then the predicted values ​​are... The above is the abnormal data processing module of the entire model, replacing the original abnormal data.

[0082] Neural Network Module:

[0083] Time series data of factors that show significant changes in the fire process are obtained from known datasets, including temperature, carbon monoxide concentration and smoke concentration, and feature values ​​that can accurately reflect the data change process are extracted from them.

[0084] The input data is ,express Sensor data in The value of the moment. Indicates the number of different types of sensors. The dataset consists of time-labeled sequences;

[0085] Information extraction is performed, including: moving average, baseline error, forecast baseline error, and comprehensive change index. The moving average is a well-known technical indicator used to predict future data in time series analysis. It enhances the judgment of data changes. The formula for the moving average is:

[0086]

[0087] express Sensor data in Moving average over time;

[0088] The moving average is a key element in extracting changes in a time series. Based on this, the baseline error and the baseline error in the prediction stage can be calculated. The baseline error is calculated as the difference between the current sensor value and the environmental baseline value. This directly reflects the degree of deviation between the current data and the environmental baseline. The environmental baseline value is represented by the moving average. The baseline error formula is expressed as:

[0089]

[0090] express Sensor data in At any time and The difference is used as an input to the neural network.

[0091] To better predict fire occurrences in advance, the model determines the error by comparing the predicted data for the next stage with a set baseline. The baseline error is then calculated using the exponential smoothing formula:

[0092]

[0093] in and The two constants are not directly related; they are used to weight historical and current values. express Sensor data in The model then extracts the cyclical trend of the sensor's previous stage reference error, the previous stage reference error, and the reference error of the prediction stage, and utilizes this trend. Returns only the characteristics of -1, 0, and 1, and returns the cyclic trend indicators for different sensors:

[0094]

[0095] The cyclical trend indicator is the baseline error of the previous period, the baseline error, and the moving baseline error of the forecast period. The summation operation under the function, the cyclic trend indicator is represented as:

[0096]

[0097] express Sensor data in The comprehensive moving reference error at any given moment is a comprehensive reflection of the degree of change of the same sensor at the current moment;

[0098] Since the model employs a backpropagation (BP) neural network, which uses a sigmoid function as the activation function, to prevent smaller values ​​in the output data from being swallowed up by larger values, the outlier-handled data needs to be processed. and the information extracted , and Normalization is performed, and the specific normalization formula is as follows:

[0099]

[0100] in For the input vector, This represents the normalized value of the parameter input. This represents the minimum value of the input vector. This represents the maximum value of the input vector;

[0101] The model will , , , The normalized data is used as the input to the neural network, and the fire probability y is used as the output of the neural network.

[0102] The model treats the input to the neural network as:

[0103] , respectively represent the first The data collection of temperature, CO concentration, smoke concentration, temperature baseline error, CO concentration baseline error, smoke concentration baseline error, temperature prediction baseline error, CO concentration prediction baseline error, smoke concentration prediction baseline error, temperature cycle trend index, CO concentration cycle trend index, and smoke concentration cycle trend index is beneficial for model formation.

[0104] The algorithm flow of the robust fire early warning system based on multi-source time series data fusion includes initializing the network, using the input data as the input vector of the BP neural network linear model, with each set of input data, i.e., each row, corresponding to a fire probability label, and determining the network structure as a three-layer structure: input layer, hidden layer, and output layer, to facilitate subsequent operations.

[0105] The following is information for each layer:

[0106] Expected output Indicates the first The expected output of the group is the fire probability;

[0107] The input to the hidden layer is: ;

[0108] The output of the hidden layer is: ;

[0109] The input to the output layer is: ;

[0110] The output of the output layer is: ;

[0111] The weights of the input layer and the hidden layer are The threshold is ;

[0112] Weights of hidden layers and output layers The threshold is ;

[0113] Randomly select a set of samples as input and target samples: the first Group input and the corresponding expected output Select error function Set the calculation precision and number of learning iterations;

[0114] Forward propagation: The input data is passed through the network, the output of each neuron is calculated, and the weighted input and the output of the activation function are calculated in each layer;

[0115] Calculate the input of each unit in the hidden layer:

[0116]

[0117] Output of the hidden layer:

[0118]

[0119] Inputs to each neuron in the output layer:

[0120]

[0121] Outputs of each neuron in the output layer:

[0122] ;

[0123] Error calculation: The output of the neural network is compared with the actual target, and the error is calculated using an error function;

[0124] Calculate the unit error between the actual output layer and the target output layer:

[0125]

[0126] Calculate the error of each unit in the intermediate layer:

[0127] ;

[0128] Backpropagation: Starting from the output layer, calculate the error gradient for each layer.

[0129] The chain rule is used to calculate the contribution of each parameter to the error, and the error gradient is propagated backward to each layer to update each parameter to minimize the error:

[0130] Adjust the weights and thresholds from the output layer to the hidden layer:

[0131]

[0132]

[0133] Correct the weights and thresholds from the hidden layer to the input layer:

[0134]

[0135]

[0136] in, For learning rate, ;

[0137] Update iteration: Determine the number of training iterations based on the loss of the training data, and repeat forward propagation and backward propagation until the number of learning iterations reaches the preset value;

[0138] The model was trained and simulated using Python. The National Institute of Standards and Technology (NIST) public dataset was selected as the training sample. The learning rate was set to 0.001, the training batch size was set to 32, and the iteration was repeated 1000 times to obtain the final model.

[0139] To address the problems of low accuracy, high false alarm rate, and poor robustness in traditional fire early warning methods, this invention proposes a robust fire early warning system based on multi-source time-series data fusion. This model utilizes the time-series characteristics of sensor data to train a model with high accuracy, low false alarm rate, and low false alarm rate through a BP neural network. At the same time, it estimates and replaces abnormal data collected by sensors, effectively improving the robustness of fire early warning.

[0140] First, anomaly detection is performed based on time-series data collected by temperature, CO concentration, and smoke concentration sensors. If the data value for a certain time period is outside the set confidence interval, an interpolation algorithm is used to estimate the value for that time period to replace the outlier. Then, the measurement data from the three sensors, the baseline error, the prediction baseline error, and the cyclic trend index are used as inputs to a neural network, along with the fire probability as a label for training. The trained model ultimately achieves the goal of accurately predicting whether a fire will occur. Based on temperature, CO concentration, and smoke concentration data collected from multiple sensors, and by fully utilizing the temporal evolution patterns of sensor data and preprocessing methods for anomalies, a robust and high-performance multi-source time-series data fusion model is proposed. The key elements of an accurate fire early warning model with a low false alarm rate are as follows: For data such as temperature, CO concentration, and smoke concentration collected by sensors, a baseline value is set for each type of data. Then, based on the collected data, the difference between each time point and the baseline value (baseline error), the difference between the predicted data at the next time point and the baseline value (prediction baseline error), and their cyclical trend indicators are calculated to reflect the relevant characteristics of this type of data at the current moment. These are then used as input data for training a BP neural network. For cases where abnormal data may appear in the temperature, CO concentration, and smoke concentration collected by sensors, an autoregressive function is used to predict and replace the original abnormal data, thereby reducing the probability of false alarms in fire early warning.

[0141] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A robust fire early warning system based on multi-source time-series data fusion, characterized in that, include Data anomaly handling module: The fire early warning system includes three types of sensors in its hardware: temperature sensor, CO concentration sensor, and smoke concentration sensor. The time-series data collected at a specific point in time is used... It means that among them Represents the type of sensor. Represents a time series; First, anomaly detection is performed on the collected time-series data to determine... Is it within the confidence interval? ； These are interval values ​​set according to actual conditions. If it exceeds this range, it means The data contains anomalies and requires anomaly handling. Experiments with missing data have shown that autoregressive models are effective for predicting unknown data. Therefore, the autoregressive formula is used to predict outlier values: ； In the autoregressive formula, This is a random error value. The autocorrelation coefficient represents the relationship between the current observation and past data. The coefficients of the linear relationship between the observations at each time point are used to predict the data at the anomalous time points, and then the predicted values ​​are... The above is the abnormal data processing module of the entire model, replacing the original abnormal data. Neural Network Module: Time series data of factors that show significant changes in the fire process are obtained from known datasets, including temperature, carbon monoxide concentration and smoke concentration, and feature values ​​that can accurately reflect the data change process are extracted from them. The input data is ,express Sensor data in The value of the moment. Indicates the number of different types of sensors. The dataset consists of time-labeled sequences; Information extraction is performed, including: moving average, baseline error, forecast baseline error, and cyclic trend indicator. The moving average is a well-known technical indicator used to predict future data in time series analysis. It enhances the judgment of data changes. The formula for the moving average is: ； express Sensor data in Moving average over time; The moving average is a key element in extracting changes in a time series. Based on this, the baseline error and the baseline error in the prediction stage can be calculated. The baseline error is calculated as the difference between the current sensor value and the environmental baseline value. This directly reflects the degree of deviation between the current data and the environmental baseline. The environmental baseline value is represented by the moving average. The baseline error formula is expressed as: ； express Sensor data in At any time and The difference is used as an input to the neural network. To better predict fire occurrences in advance, the model determines the error by comparing the predicted data for the next moment with a set baseline. The baseline error is then calculated using the exponential smoothing formula: ； in and The two constants are not directly related; they are used to weight historical and current values. express Sensor data in The model then extracts the baseline error during the time prediction phase, and subsequently performs a cyclical trend extraction on the sensor's baseline error from the previous time step, the current time step, and the predicted baseline error for the next time step. Returns only the characteristics of -1, 0, and 1, and returns the cyclic trend index of the same sensor: ； The cyclic trend indicator is the baseline error of the previous time step, the baseline error of the current time step, and the baseline error of the prediction phase. The summation operation under the function, the cyclic trend indicator is represented as: ； express Sensor data in The cyclical trend indicator at any given moment is a comprehensive reflection of the degree of trend change of the same sensor at the previous moment, the current moment, and the next moment; Since the model employs a backpropagation (BP) neural network, which uses a sigmoid function as the activation function, to prevent smaller values ​​in the output data from being swallowed up by larger values, the outlier-handled data needs to be processed. and the information extracted , and Normalization is performed, and the specific normalization formula is as follows: ； in For the input vector, This represents the normalized value of the parameter input. This represents the minimum value of the input vector. This represents the maximum value of the input vector; The model will , , , The normalized data is used as the input to the neural network, and the fire probability y is used as the output of the neural network.

2. A robust fire early warning system based on multi-source time-series data fusion according to claim 1, characterized in that, The model treats the input of the neural network as: , respectively represent the first Group temperature, CO concentration, smoke concentration, temperature baseline error, CO concentration baseline error, smoke concentration baseline error, temperature prediction baseline error, CO concentration prediction baseline error, smoke concentration prediction baseline error, temperature cycle trend index, CO concentration cycle trend index, smoke concentration cycle trend index.

3. A robust fire early warning system based on multi-source time-series data fusion according to claim 1, characterized in that, The algorithm flow of the robust fire early warning system based on multi-source time series data fusion includes initializing the network, using the input data as the input vector of the BP neural network linear model, with each set of input data, i.e., each row, corresponding to a fire probability label, and determining the network structure as a three-layer structure: input layer, hidden layer, and output layer.

4. A robust fire early warning system based on multi-source time-series data fusion according to claim 3, characterized in that, The following is information for each layer: Expected output Indicates the first The expected output of the group is the fire probability; The input to the hidden layer is: ; The output of the hidden layer is: ; The input to the output layer is: ; The output of the output layer is: ; The weights of the input layer and the hidden layer are The threshold is ; Weights of hidden layers and output layers The threshold is ; Randomly select a set of samples as input and target samples: the first Group input and the corresponding expected output Select error function Set the calculation precision and the number of learning iterations.

5. A robust fire early warning system based on multi-source time-series data fusion according to claim 4, characterized in that, Forward propagation: The input data is passed through the network, the output of each neuron is calculated, and the weighted input and the output of the activation function are calculated in each layer; Calculate the input of each unit in the hidden layer: ； Output of the hidden layer: ； Inputs to each neuron in the output layer: ； Outputs of each neuron in the output layer: 。 6. A robust fire early warning system based on multi-source time-series data fusion according to claim 5, characterized in that, Error calculation: The output of the neural network is compared with the actual target, and the error is calculated using an error function; Calculate the unit error between the actual output layer and the target output layer: ； Calculate the error of each unit in the intermediate layer: 。 7. A robust fire early warning system based on multi-source time-series data fusion according to claim 6, characterized in that, Backpropagation: Starting from the output layer, calculate the error gradient for each layer. The chain rule is used to calculate the contribution of each parameter to the error, and the error gradient is propagated backward to each layer to update each parameter to minimize the error: Adjust the weights and thresholds from the output layer to the hidden layer: ； ； Correct the weights and thresholds from the hidden layer to the input layer: ； ； in, For learning rate, .

8. A robust fire early warning system based on multi-source time-series data fusion according to claim 7, characterized in that, Update iteration: Determine the number of training iterations based on the loss of the training data, and repeatedly perform forward propagation and backward propagation until the number of learning iterations reaches the preset value.

9. A robust fire early warning system based on multi-source time-series data fusion according to claim 8, characterized in that, The model was trained and simulated using Python. The National Institute of Standards and Technology (NIST) public dataset was selected as the training sample, with a learning rate of 0.001, a training batch size of 32, and 1000 iterations to obtain the final model.

Images