Air pollution control facility management and control method based on hidden space sLSTM

Abstract

The embodiment of the invention provides an air pollution control facility management and control method based on hidden space sLSTM. The method is applied to pollution treatment. The method comprises the following steps: preprocessing collected operation data of the atmospheric pollution control facility to obtain multi-dimensional time sequence data; inputting the multi-dimensional time sequence data into a hidden space sLSTM model, executing hidden space embedding, discrete implicit vector mapping, time sequence prediction, prediction sequence reconstruction and abnormal sequence detection operation by the hidden space sLSTM model, and outputting an abnormal sequence; and managing and controlling the corresponding atmospheric pollution control facilities according to the operation data of the atmospheric pollution control facilities in the abnormal sequence. In this way, the complex time sequence dependency relationship between the operation data of the pollution treatment facilities is obtained, and the real-time performance, accuracy and efficiency of pollution treatment facility management and control are improved.

Description

Technical Field

[0001] The present disclosure relates to the field of pollution control, and in particular, to a method for controlling an air pollution control facility based on a hidden space sLSTM. Background Art

[0002] With the rapid development of industrialization and urbanization, the global air pollution problem has become increasingly severe, and it has become a major challenge affecting human health, the ecosystem, and the sustainable development of social economy. As one of the treatment means, air pollution control facilities play an irreplaceable role in restricting pollutant emissions and improving air quality. These facilities separate, transform, or purify pollutants through complex technical means, and have important strategic significance in improving air quality, protecting human health, and maintaining environmental balance. Therefore, real-time status monitoring of air pollution control facilities to ensure their stable and efficient operation not only helps to improve the treatment effect, but also can effectively reduce the emission risk.

[0003] Traditional methods for controlling air pollution control facilities mostly rely on regular maintenance and manual inspections. However, this method appears to be relatively lagging in the modern complex industrial environment, and there are problems such as high cost and low efficiency, and it cannot meet the requirements of real-time control. With the progress of technology, the traditional "human defense" mode has changed to be dominated by "technical defense", using high-precision sensors in air pollution control facilities to record equipment status information in real time. A series of time series formed by these multi-dimensional data provide the possibility for the comprehensive perception of equipment status.

[0004] Current methods for controlling air pollution control facilities mainly include methods based on preset threshold comparison, methods based on image processing, and methods based on traditional machine learning algorithms. These methods have the following disadvantages: The method based on preset threshold comparison is difficult to achieve precise control; The method based on image processing depends on the appearance characteristics of the equipment, and it is difficult to capture potential faults inside the equipment, and the computational resource requirements are high, making it difficult to meet the real-time requirements of industrial equipment monitoring; The method based on traditional machine learning algorithms cannot comprehensively capture potential patterns in the data, nor can it adapt to new data or environmental changes in a timely manner, and its performance in dynamic industrial scenarios is poor, and it is not suitable for monitoring under large-scale data. Summary of the Invention

[0005] The present disclosure provides a method, device, equipment, and storage medium for controlling an air pollution control facility based on a hidden space sLSTM.

[0006] According to the first aspect of the present disclosure, there is provided a method for controlling an air pollution control facility based on a hidden space sLSTM. The method includes:

[0007] Preprocess the operation data of the collected air pollution control facility to obtain multi-dimensional time series data;

[0008] Input the multi-dimensional time series data into the hidden space sLSTM model so that the hidden space sLSTM model performs operations of hidden space embedding, discrete hidden vector mapping, time series prediction, predicted sequence reconstruction, and abnormal sequence detection, and outputs the abnormal sequence;

[0009] Control the corresponding air pollution control facilities according to the operation data of the air pollution control facilities in the abnormal sequence.

[0010] In some realizable ways of the first aspect, the preprocessing includes:

[0011] Data cleaning, noise and outlier removal, standardization and normalization processing.

[0012] In some realizable ways of the first aspect, the hidden space embedding is implemented by the following method:

[0013] Divide the multi-dimensional time series data into multiple time windows with a preset length by the hidden space sLSTM model, and map the time windows to a low-dimensional hidden space to obtain low-dimensional hidden vectors; wherein, the multiple time windows with a preset length form a time window sequence.

[0014] In some realizable ways of the first aspect, the discrete hidden vector mapping is implemented by the following method:

[0015] Quantize the low-dimensional hidden space in the low-dimensional hidden space according to the encoding dictionary in the hidden space sLSTM model to obtain discrete hidden vectors.

[0016] In some realizable ways of the first aspect, the time series prediction is implemented by the following method:

[0017] Transmit the discrete hidden vectors in the discrete hidden space to the sLSTM module in the hidden space sLSTM model, and obtain the predicted time series according to the hybrid memory mechanism and exponential gating mechanism adopted by the sLSTM module.

[0018] In some realizable ways of the first aspect, the hybrid memory mechanism includes:

[0019] Process the discrete hidden vectors through multiple parallel processing units in the sLSTM module to obtain time series dependencies; wherein, each parallel processing unit has its own calculation path; wherein, the calculation paths respectively include the calculation paths of their own input gates, forget gates, output gates, and sequence states.

[0020] In some realizable ways of the first aspect, the exponential gating mechanism includes:

[0021] Separate gate control mechanisms are set in the input gate, forget gate, and output gate of each parallel processing unit for the discrete hidden vector, and the gate control mechanisms are adjusted so that the outputs of the input gate and forget gate of each parallel processing unit are greater than 1.

[0022] In some realizable ways of the first aspect, the prediction sequence reconstruction is achieved by the following method:

[0023] According to the decoder in the latent space sLSTM model, the predicted time series is reconstructed to generate a reconstructed time window sequence corresponding to the predicted time series, and the reconstructed time window sequence is mapped back to the high-dimensional space where the time window sequence is located.

[0024] In some realizable ways of the first aspect, the abnormal sequence detection operation is achieved by the following method:

[0025] The reconstructed time window sequence in the high-dimensional space is compared with the time window sequence through the latent space sLSTM model, and the anomaly score of the time window sequence is calculated using the anomaly score function;

[0026] The time window sequence corresponding to the anomaly score greater than the preset threshold is output as the abnormal sequence.

[0027] According to the second aspect of the present disclosure, there is provided a control device for an air pollution control facility based on a latent space sLSTM. The device includes:

[0028] A multi-dimensional time series data acquisition module for preprocessing the operation data of the collected air pollution control facility to obtain multi-dimensional time series data;

[0029] An abnormal sequence acquisition module for inputting the multi-dimensional time series data into the latent space sLSTM model so that the latent space sLSTM model performs latent space embedding, discrete hidden vector mapping, time series prediction, prediction sequence reconstruction, and abnormal sequence detection operations, and outputs an abnormal sequence;

[0030] A control module for the control facility for controlling the corresponding air pollution control facility according to the operation data of the air pollution control facility in the abnormal sequence.

[0031] According to the third aspect of the present disclosure, there is provided an electronic device. The electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method as described above.

[0032] According to the fourth aspect of the present disclosure, there is provided a non-transitory computer-readable storage medium storing computer instructions for causing a computer to execute the method as described above.

[0033] In the present disclosure, the operation data of the air pollution control facilities collected are preprocessed to obtain multi-dimensional time series data; the multi-dimensional time series data is input into the hidden space sLSTM model so that the hidden space sLSTM model performs operations of hidden space embedding, discrete hidden vector mapping, time series prediction, prediction sequence reconstruction, and abnormal sequence detection, and outputs an abnormal sequence; according to the operation data of the air pollution control facilities in the abnormal sequence, the corresponding air pollution control facilities are controlled. In this way, the complex time series dependence relationship between the operation data of the pollution control facilities is obtained, and the real-time performance, accuracy, and efficiency of the control of the pollution control facilities are improved.

[0034] It should be understood that the content described in the summary of the invention section is not intended to limit the key or important features of the embodiments of the present disclosure, nor is it used to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. Brief Description of the Drawings

[0035] In combination with the drawings and with reference to the following detailed description, the above and other features, advantages, and aspects of the embodiments of the present disclosure will become more obvious. The drawings are used to better understand the solution and do not constitute a limitation to the present disclosure. In the drawings, the same or similar reference numerals represent the same or similar elements, where:

[0036] Figure 1 shows a flowchart of a method for controlling air pollution control facilities based on a hidden space sLSTM provided by an embodiment of the present disclosure;

[0037] Figure 2 shows a schematic diagram of a layer normalization residual structure provided by an embodiment of the present disclosure;

[0038] Figure 3 shows a structural diagram of a device for controlling air pollution control facilities based on a hidden space sLSTM provided by an embodiment of the present disclosure;

[0039] Figure 4 shows a structural diagram of an exemplary electronic device capable of implementing the embodiments of the present disclosure. Detailed Description of the Embodiments

[0040] To make the objectives, technical solutions, and advantages of the embodiments of the present disclosure clearer, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Apparently, the described embodiments are some, but not all, of the embodiments of the present disclosure. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present disclosure without creative efforts shall fall within the protection scope of the present disclosure.

[0041] In addition, the term "and / or" in this text is merely a description of the association relationship between associated objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in this text generally represents an "or" relationship between the preceding and following associated objects.

[0042] In view of the problems in the background art, the embodiments of the present disclosure provide a control method for air pollution control facilities based on a hidden space sLSTM. Specifically, the operation data of the air pollution control facilities collected is preprocessed to obtain multi-dimensional time series data; the multi-dimensional time series data is input into the hidden space sLSTM model so that the hidden space sLSTM model performs operations such as hidden space embedding, discrete hidden vector mapping, time series prediction, prediction sequence reconstruction, and abnormal sequence detection, and outputs an abnormal sequence; according to the operation data of the air pollution control facilities in the abnormal sequence, the corresponding air pollution control facilities are controlled. In this way, the complex time series dependence relationship between the operation data of the pollution control facilities is obtained, and the real-time performance, accuracy, and efficiency of the control of the pollution control facilities are improved.

[0043] The following will, with reference to the accompanying drawings, through specific embodiments, elaborate in detail on the control method and device for air pollution control facilities based on a hidden space sLSTM provided by the embodiments of the present disclosure.

[0044] Figure 1 FIG. shows a flowchart of a control method for air pollution control facilities based on a hidden space sLSTM provided by the embodiments of the present disclosure. Method 100 includes the following steps:

[0045] S110, preprocess the operation data of the air pollution control facilities collected to obtain multi-dimensional time series data.

[0046] In some embodiments, the operation data of the air pollution control facilities is collected in real time by using Internet of Things sensing devices such as differential pressure gauges, flow meters, and temperature sensors installed on the air pollution control facilities. Among them, the operation data includes the pressure difference of the filtration and adsorption medium, the exhaust gas flow rate, and the temperature of the activated carbon, etc. These operation data have multi-dimensional time series characteristics, and each dimension represents a sequence of different physical quantities changing over time, covering the timestamps and values collected synchronously by multiple sensors.

[0047] In some embodiments, the preprocessing includes:

[0048] Data cleaning, noise and outlier removal, standardization and normalization processing.

[0049] In some embodiments, the multi-dimensional time series data X can be expressed as X = {x1, x2,..., x N}, where x u ∈Rm , which represents an m-dimensional vector at the u-th time point and contains multimodal information collected from m sensors.

[0050] S120, input the multi-dimensional time series data into the latent space sLSTM model, so that the latent space sLSTM model performs latent space embedding, discrete latent vector mapping, time series prediction, predicted sequence reconstruction and abnormal sequence detection operations, and outputs an abnormal sequence.

[0051] In some embodiments, the latent space sLSTM model includes an encoder, a decoder, and an sLSTM module.

[0052] In some embodiments, the latent space embedding is achieved by the following method:

[0053] Divide the multi-dimensional time series data into multiple time windows of a preset length through the latent space sLSTM model, and map the time windows to a low-dimensional latent space to obtain low-dimensional latent vectors; wherein, the multiple time windows of the preset length form a time window sequence;

[0054] Further, using the window sliding method, divide the multi-dimensional time series data into multiple time windows of a preset length, each time window contains data of p time points, p represents the preset length, these time windows can capture the dynamic changes of the running data in the short term, and then map each time window to a low-dimensional latent space through the encoder of the latent space sLSTM model to obtain a low-dimensional latent vector; wherein, the time window w at time point t t can be expressed as w t =[x t-p+1 , x t-p+2 ,..., x t , the low-dimensional latent vector corresponding to the time window w at time point t t can be expressed as e′ t , the time window sequence W at time point t t can be expressed as W t =[w t-(k-1)×p , w t-(k-2)×p ,…,w t , k is the number of windows.

[0055] In some embodiments, the discrete latent vector mapping is achieved by the following method:

[0056] Quantify the low-dimensional latent vectors in the low-dimensional latent space according to the encoding dictionary in the latent space sLSTM model to obtain discrete latent vectors.

[0057] At the beginning of the training of the latent space sLSTM model in some embodiments, the encoding dictionary V is randomly initialized, where the encoding dictionary can be expressed as v lis the l-th discrete vector of dimension n in the coding dictionary, and the length of the coding dictionary is L;

[0058] During the process of initializing the coding dictionary, the discrete vector v l is randomly generated and continuously updated during subsequent training to capture the characteristic patterns of multi-dimensional time series data in the discrete latent space;

[0059] Furthermore, the low-dimensional latent vector e′ t is quantized (i.e., discretized) in the following manner to obtain the discrete latent vector e t :

[0060]

[0061] where ‖·‖2 represents the Euclidean distance.

[0062] In some embodiments, the time window sequence W at the t-th time point t The discrete latent vector sequence obtained after discretization processing can be expressed as where represents the discrete latent vector corresponding to the j-th time window in the window sequence W t in.

[0063] In some embodiments, the discrete latent vector mapping operation can simplify data representation and help improve the computational efficiency and generalization ability of the latent space sLSTM model.

[0064] In some embodiments, time series prediction is achieved through the following method:

[0065] The discrete latent vectors in the discrete latent space are transmitted to the sLSTM module in the latent space sLSTM model, and according to the hybrid memory mechanism and exponential gating mechanism adopted by the sLSTM module, the predicted time series is obtained.

[0066] In some embodiments, the hybrid memory mechanism includes:

[0067] The discrete latent vectors are processed by multiple parallel processing units in the sLSTM module to obtain the temporal dependence relationship; among them,

[0068] Each parallel processing unit has its own computational path; among them, the computational paths respectively include the computational paths of their own input gates, forget gates, output gates, and sequence states.

[0069] In some embodiments, the exponential gating mechanism includes:

[0070] Separate gating mechanisms are set for the input gate, forget gate, and output gate of the discrete hidden vector in each parallel processing unit, and the gating mechanisms are adjusted so that the outputs of the input gate and forget gate of each parallel processing unit are greater than 1.

[0071] In some embodiments, the sLSTM module is embedded in a layer normalization residual structure, and this layer normalization residual structure is as Figure 2 shown. The data input to the layer normalization residual structure first passes through a layer normalization module LN for processing, and then is divided into two paths for output; the data of the first path of output passes through a convolutional layer Conv4 with a window size of 4 and a Swish activation function in sequence and then outputs two paths of data again. These two paths of data are respectively input into two parallel processing units for calculating the inputs of the input gate i and the forget gate f; the data of the second path of output is divided into two paths of data and respectively input into another two parallel processing units for processing to calculate the inputs of the cell input z and the output gate o; the inputs of the above input gate, forget gate, cell input, and output gate are updated for the sequence state through the sLSTM module, and the output of the sLSTM module is group-normalized through a group normalization module GN to ensure data stability; the output of GN is divided into two paths of data, and dimensionality increase operations are respectively performed. The projection factor PF for the dimensionality increase operation is 4 / 3. After the dimensionality increase operation, one path of data passes through a GELU activation function for processing and is multiplied element-wise with the other path of data. The result of the multiplication undergoes a dimensionality reduction operation, and the projection factor PF for the dimensionality reduction operation is 3 / 4. The data after the dimensionality reduction operation is projected back to the original dimensional space; finally, the data projected back to the original dimensional space and the input data of the layer normalization residual structure are added through a residual connection to obtain the final output of this layer normalization residual structure.

[0072] In some embodiments, each parallel processing unit is processed using a block diagonal layer, Figure 2 where NH in

[0073] represents the number of parallel processing units; Specifically, inside the sLSTM module, a weighted sum is performed on the discrete hidden vector j-1 and the hidden state h of the previous time window to obtain the pre-activation value of the cell input j , and the cell input z is obtained through the tanh activation function

[0074]

[0075] where, r z and b z are respectively the weight matrix corresponding to the discrete hidden vector , the weight matrix of the hidden state h of the previous time window sequence j-1 and the bias term, The pre-activation value input for the cell is the tanh activation function;

[0076] An exponential gating mechanism is introduced to enable the outputs of the input gate and the forget gate to be greater than 1, thereby allowing for more flexible information retention and update, enabling the hidden space sLSTM model to update and overwrite previously stored information more quickly. After introducing the exponential gating mechanism, the calculation formulas for the input gate, the forget gate, and the output gate are as follows:

[0077]

[0078] i j , f j , o j respectively represent the output of the input gate, the output of the forget gate, and the output of the output gate corresponding to the j-th time window, respectively represent the pre-activation values of the input gate i j , the forget gate f j , and the output gate o j , are respectively the weight matrices from the discrete hidden vector to the input gate i j , the forget gate f j , and the output gate o j , r i , r f , r o are respectively the weight matrices from the hidden state h h-1 to the input gate i j , the forget gate f j , and the output gate o j , b i , b f , b o are respectively the bias terms corresponding to the input gate i j , the forget gate f j , and the output gate o j , σ(.) is the sigmoid activation function, and exp(.) is the exponential activation function;

[0079] To keep the outputs of the input gate and the forget gate within a reasonable range, a stabilization state parameter is introduced, and its calculation formula is as follows:

[0080] K j = max(log(f j ) + K j-1 , log(i j ))

[0081] K j is the stabilization state parameter corresponding to the j-th time window, and K j-1 is the stabilization state parameter corresponding to the (j - 1)-th time window;

[0082] Scale the outputs of the input gate and the forget gate according to the stabilization state parameters to obtain the stabilized output i' of the input gate j and the stabilized output f' of the forget gate j , and the calculation formula is as follows:

[0083] i' j = exp(log(i j ) - K j )

[0084] f' j = exp(log(f j ) + K j-1 - K j )

[0085] Use the stabilized output i' of the input gate j and the stabilized output f' of the forget gate j to calculate the cell state c at the current (j-th time window) k , and the formula is as follows:

[0086] c k = f' j z j-1 + i' j z j

[0087] z j is the cell input corresponding to the j-th time window, and z j-1 is the cell input corresponding to the (j - 1)-th time window;

[0088] To further alleviate the numerical instability brought by the exponential gating mechanism, introduce a normalization state parameter to balance the growth of the cell state. The calculation formula of the normalization state parameter n corresponding to the j-th time window is as follows: j The calculation formula of the normalization state parameter n corresponding to the j-th time window is as follows:

[0089] n j = f' j n j-1 + i' j

[0090] According to the cell state c j , the normalization state parameter n j and the output gate output o j , calculate the hidden state h j , and the formula is as follows:

[0091] h j = o j *(c j / n j )

[0092] Discrete hidden vector The final hidden state at the current time point obtained through the layer normalization residual structure is:

[0093]

[0094] GN(.) is the group normalization operation, and MLP(.) is the multi-layer perceptron operation;

[0095] According to the final hidden state at the current time point and the discrete hidden vector calculate the discrete hidden vector at the next time point The formula is as follows:

[0096]

[0097] Therefore, the sLSTM module can predict the discrete hidden vector at the next time point based on the final hidden state and the discrete hidden vector at the current time point, and repeat this process in a loop to finally obtain j predicted discrete hidden vectors.

[0098] In some embodiments, the prediction loss function L used during the training process of the sLSTM module pred is:

[0099]

[0100] represents the discrete hidden vector predicted for the j-th time window in the discrete hidden vector E t corresponding to the predicted time series.

[0101] In some embodiments, the layer normalization residual structure helps the gradient to effectively propagate in the deep network by providing a direct connection path, alleviates the problem of gradient vanishing, and enables more stable model training.

[0102] In some embodiments, the prediction sequence reconstruction is achieved through the following method:

[0103] According to the decoder in the latent space sLSTM model, reconstruct the predicted time series to generate a reconstructed time window sequence corresponding to the predicted time series, and map the reconstructed time window sequence back to the high-dimensional space where the time window sequence is located.

[0104] In some embodiments, the non-linear transformation layer of the decoder maps the reconstructed time window sequence back to the high-dimensional space where the time window sequence is located.

[0105] In some embodiments, during the training process of the encoder and decoder in the latent space sLSTM model, the loss function L total includes minimizing the reconstruction loss function L recon, quantization loss function \(L\) VQ , commitment loss function \(L\) commit and regularization loss function \(L\) reg ; among them, the regularization loss function is used to prevent the model from overfitting;

[0106] Further, \(L\) total = \(L\) recon + \(\beta L\) VQ + \(\gamma L\) commit + \(\lambda L\) reg , where \(\beta\), \(\gamma\), and \(\lambda\) are the weights corresponding to the quantization loss function, commitment loss function, and regularization loss function respectively, and these weights are used to balance the performance of the encoder and decoder in terms of reconstruction accuracy, quantization accuracy, output stability, and prevention of overfitting;

[0107] Among them, is the \(t\)-th time window sequence of the reconstruction. Minimizing the reconstruction loss function represents the difference between the reconstructed time window sequence and the time window sequence; \(sg[.]\) represents the stop gradient function, which is used to prevent backpropagation to ensure that only the parts of the encoder or the coding dictionary that need to be updated are adjusted during the training process. These two loss functions are used to update the discrete latent vector to ensure the stability and consistency of the encoder during the quantization process; \(\Theta\) represents a preset threshold, is the second norm function.

[0108] In some embodiments, the abnormal sequence detection operation is implemented by the following method:

[0109] Compare the reconstructed time window sequence in the high-dimensional space with the time window sequence through the latent space sLSTM model, and calculate the abnormal score of the time window sequence using the abnormal score function;

[0110] Output the time window sequence corresponding to the abnormal score greater than the preset threshold as the abnormal sequence.

[0111] In some embodiments, the abnormal score function \(d\) t is as follows:

[0112]

[0113] Among them, \(k\) represents the number of windows used in the window sliding method.

[0114] In some embodiments, the preset threshold is determined by the following method:

[0115] When using the validation set to verify the latent space sLSTM model, take the maximum value of the F1 score as the preset threshold;

[0116] Among them, the F1 score is the harmonic mean of Precision and Recall, and the calculation formula is as follows:

[0117]

[0118] Precision represents the ratio of the true abnormal sequences detected by the latent space sLSTM model to the detected abnormal sequences, and Recall represents the proportion of the true abnormal sequences correctly detected by the latent space sLSTM model.

[0119] S130. According to the operation data of the air pollution control facilities in the abnormal sequence, control the corresponding air pollution control facilities.

[0120] According to an embodiment of the present disclosure, preprocess the collected operation data of the air pollution control facilities to obtain multi-dimensional time series data; input the multi-dimensional time series data into the latent space sLSTM model so that the latent space sLSTM model performs latent space embedding, discrete latent vector mapping, time series prediction, prediction sequence reconstruction, and abnormal sequence detection operations, and output abnormal sequences; according to the operation data of the air pollution control facilities in the abnormal sequence, control the corresponding air pollution control facilities. In this way, obtain the complex time series dependence relationship between the operation data of the pollution control facilities, and improve the real-time performance, accuracy and efficiency of the pollution control facility control.

[0121] It should be noted that for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that the present disclosure is not limited by the described action sequence, because according to the present disclosure, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to the present disclosure.

[0122] The above is the introduction of the method embodiments. The following further illustrates the solution of the present disclosure through device embodiments.

[0123] Figure 3 Fig. shows the structural diagram of an air pollution control facility control device based on the latent space sLSTM provided by an embodiment of the present disclosure. The device 300 includes:

[0124] A multi-dimensional time series data acquisition module 310, configured to preprocess the collected operation data of the air pollution control facilities to obtain multi-dimensional time series data.

[0125] In some embodiments, the multi-dimensional time series data acquisition module 310 is specifically configured to:

[0126] The preprocessing includes:

[0127] Data cleaning, noise and outlier removal, standardization and normalization processing.

[0128] Anomaly sequence acquisition module 320, configured to input the multi-dimensional time series data into the latent space sLSTM model, so that the latent space sLSTM model performs latent space embedding, discrete latent vector mapping, time series prediction, predicted sequence reconstruction, and anomaly sequence detection operations, and outputs an anomaly sequence.

[0129] In some embodiments, the anomaly sequence acquisition module 320 is specifically configured to:

[0130] The latent space embedding is implemented by the following method:

[0131] The multi-dimensional time series data is divided into multiple time windows of a preset length by the latent space sLSTM model, and the time windows are mapped to a low-dimensional latent space to obtain low-dimensional latent vectors; wherein, the multiple time windows of the preset length form a time window sequence.

[0132] In some embodiments, the anomaly sequence acquisition module 320 is specifically further configured to:

[0133] The discrete latent vector mapping is implemented by the following method:

[0134] According to the encoding dictionary in the latent space sLSTM model, the low-dimensional latent space in the low-dimensional latent space is quantized to obtain discrete latent vectors.

[0135] In some embodiments, the anomaly sequence acquisition module 320 is specifically further configured to:

[0136] The time series prediction is implemented by the following method:

[0137] The discrete latent vectors in the discrete latent space are transmitted to the sLSTM module in the latent space sLSTM model, and according to the hybrid memory mechanism and exponential gating mechanism adopted by the sLSTM module, a predicted time series is obtained.

[0138] In some embodiments, the anomaly sequence acquisition module 320 is specifically further configured to:

[0139] The hybrid memory mechanism includes:

[0140] The discrete latent vectors are processed by multiple parallel processing units in the sLSTM module to obtain temporal dependence relationships; wherein, each parallel processing unit has its own computational path; wherein, the computational paths respectively include the computational paths of their respective input gates, forget gates, output gates, and sequence states.

[0141] In some embodiments, the anomaly sequence acquisition module 320 is specifically further configured to:

[0142] An exponential gating mechanism, including:

[0143] Respectively set the gating mechanisms in the input gate, forget gate, and output gate of each parallel processing unit for the discrete hidden vector, and adjust the gating mechanisms so that the outputs of the input gate and forget gate of each parallel processing unit are greater than 1.

[0144] In some embodiments, the abnormal sequence acquisition module 320 is further specifically configured to:

[0145] The prediction sequence reconstruction is implemented by the following method:

[0146] According to the decoder in the latent space sLSTM model, reconstruct the predicted time series, generate a reconstructed time window sequence corresponding to the predicted time series, and map the reconstructed time window sequence back to the high-dimensional space where the time window sequence is located.

[0147] In some embodiments, the abnormal sequence acquisition module 320 is further specifically configured to:

[0148] The abnormal sequence detection operation is implemented by the following method:

[0149] Compare the reconstructed time window sequence in the high-dimensional space with the time window sequence through the latent space sLSTM model, and calculate the anomaly score of the time window sequence by using an anomaly scoring function;

[0150] Output the time window sequence corresponding to the anomaly score greater than a preset threshold as an abnormal sequence.

[0151] The governance facility control module 330 is configured to control the corresponding air pollution governance facility according to the operation data of the air pollution governance facility in the abnormal sequence.

[0152] It can be understood that Figure 3 Each module / unit in the device 300 shown has the function of implementing each step in the method 100 provided in the embodiments of the present disclosure and can achieve its corresponding technical effects. For the sake of brevity, they will not be described in detail here.

[0153] Figure 4 The structure diagram of an exemplary electronic device capable of implementing the embodiments of the present disclosure is shown. The electronic device 400 is intended to represent various forms of digital computers, such as, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device 400 can also represent various forms of mobile devices, such as, personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described herein and / or claimed.

[0154] As Figure 4 shown, the electronic device 400 includes a computing unit 401, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 402 or a computer program loaded from a storage unit 408 into a random access memory (RAM) 403. In the RAM 403, various programs and data required for the operation of the electronic device 400 can also be stored. The computing unit 401, the ROM 402, and the RAM 403 are connected to each other via a bus 404. An I / O interface 405 is also connected to the bus 404.

[0155] Multiple components in the electronic device 400 are connected to the I / O interface 405, including: an input unit 406, such as a keyboard, a mouse, etc.; an output unit 407, such as various types of displays, speakers, etc.; a storage unit 408, such as a magnetic disk, an optical disc, etc.; and a communication unit 409, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 409 allows the electronic device 300 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0156] The computing unit 401 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 401 include but are not limited to a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The computing unit 401 executes the various methods and processes described above, such as method 100. For example, in some embodiments, method 100 can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as the storage unit 408. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device 400 via the ROM 402 and / or the communication unit 409. When the computer program is loaded into the RAM 403 and executed by the computing unit 401, one or more steps of method 100 described above can be executed. Alternatively, in other embodiments, the computing unit 401 can be configured to execute method 100 in any other appropriate manner (e.g., by means of firmware).

[0157] The various embodiments of the systems and techniques described above in this specification can be implemented in digital electronic circuitry, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on a chip (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be a special-purpose or general-purpose programmable processor that receives data and instructions from, and transmits data and instructions to, a storage system, at least one input device, and at least one output device.

[0158] The program code for implementing the methods of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus, such that the program codes, when executed by the processor or controller, cause the functions / operations specified in the flowchart and / or block diagram to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package partially on the machine and partially on a remote machine, or entirely on the remote machine or server.

[0159] In the context of the present disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0160] It should be noted that the present disclosure also provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to execute method 100 and achieve the corresponding technical effects achieved by the method of the embodiments of the present disclosure. For the sake of brevity of description, details are not repeated herein.

[0161] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0162] The systems and techniques described herein can be implemented in a computing system including backend components (e.g., as a data server), or a computing system including middleware components (e.g., an application server), or a computing system including frontend components (e.g., a user computer having a graphical user interface or a web browser through which the user can interact with an implementation of the systems and techniques described herein), or a computing system including any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected to each other by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local area network (LAN), wide area network (WAN), and the Internet.

[0163] A computer system can include a client and a server. The client and the server are generally far from each other and typically interact through a communication network. The client-server relationship is generated by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, or a server of a distributed system, or a server incorporating a blockchain.

[0164] It should be understood that various forms of the processes shown above can be used, with steps reordered, added, or deleted. For example, the steps recited in this disclosure can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and no limitations are imposed herein.

[0165] The above specific embodiments do not constitute a limitation on the protection scope of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure shall be included within the protection scope of this disclosure.

Claims

1. A method for controlling air pollution control facilities based on latent space sLSTM, characterized in that: include: Preprocess the collected operation data of air pollution control facilities to obtain multi-dimensional time series data; Inputting the multidimensional time series data into the latent space sLSTM model, so that the latent space sLSTM model performs latent space embedding, discrete latent vector mapping, time series prediction, prediction sequence reconstruction and abnormal sequence detection operations, and outputs the abnormal sequence; According to the operation data of the air pollution control facilities in the abnormal sequence, the corresponding air pollution control facilities are controlled.

2. The method according to claim 1, characterized in that: The preprocessing comprises: Data cleaning, noise and outlier removal, standardization and normalization processing.

3. The method according to claim 1, characterized in that The latent space embedding is achieved by the following method: The multidimensional time series data is divided into multiple time windows of preset lengths through a latent space sLSTM model, and the time windows are mapped to a low-dimensional latent space to obtain a low-dimensional latent vector; wherein, multiple time windows of preset lengths constitute a time window sequence.

4. The method according to claim 3, characterized in that The discrete latent vector mapping is achieved by the following method: According to the encoding dictionary in the latent space sLSTM model, the low-dimensional latent vector in the low-dimensional latent space is quantized to obtain a discrete latent vector.

5. The method according to claim 4, characterized in that The time series prediction is achieved by the following method: The discrete latent vector in the discrete latent space is transferred to the sLSTM module in the latent space sLSTM model, and the predicted time series is obtained according to the hybrid memory mechanism and exponential gating mechanism adopted by the sLSTM module.

6. The method according to claim 5, characterized in that The hybrid memory mechanism comprises: The discrete latent vector is processed by multiple parallel processing units in the sLSTM module to obtain the temporal dependency relationship; Each parallel processing unit has its own calculation path, wherein the calculation path includes the calculation path of its own input gate, forget gate, output gate and sequence state.

7. The method according to claim 6, characterized in that The index gating mechanism comprises: A gating mechanism of the discrete latent vector in the input gate, the forget gate, and the output gate of each parallel processing unit is respectively set, and the gating mechanism is adjusted so that the output of the input gate and the forget gate of each parallel processing unit is greater than 1.

8. The method according to claim 5, characterized in that The predicted sequence reconstruction is achieved by the following method: According to the decoder in the latent space sLSTM model, the predicted time series is reconstructed to generate a reconstructed time window sequence corresponding to the predicted time series, and the reconstructed time window sequence is mapped back to the high-dimensional space where the time window sequence is located.

9. The method according to claim 8, characterized in that The abnormal sequence detection operation is implemented by the following method: The reconstructed time window sequence in high-dimensional space is compared with the time window sequence through the latent space sLSTM model, and the anomaly score function is used to calculate the anomaly score of the time window sequence; The time window sequence corresponding to the anomaly score greater than the preset threshold is output as the anomaly sequence.

10. An air pollution control facility control device based on latent space sLSTM corresponding to the method of claim 1, characterized in that: include: A multi-dimensional time series data acquisition module is used to pre-process the collected operation data of air pollution control facilities to obtain multi-dimensional time series data; An abnormal sequence acquisition module is used to input the multidimensional time series data into the latent space sLSTM model so that the latent space sLSTM model performs latent space embedding, discrete latent vector mapping, time series prediction, prediction sequence reconstruction and abnormal sequence detection operations, and outputs the abnormal sequence; The control facility management module is used to manage and control the corresponding air pollution control facilities according to the operation data of the air pollution control facilities in the abnormal sequence.

Images