Iot sensing data prediction method based on deep learning

Abstract

The application relates to a kind of deep learning-based internet of things perception data prediction method, belonging to the technical field of internet of things.The method comprises: obtaining internet of things perception long-term data and pre-processing, according to the characteristics of internet of things data with seasonality, using STL method to decompose time series data, and extracting periodicity, trend and residual characteristics.Improve the whale algorithm from the perspective of location update strategy and prevention from falling into local optimum, and use the improved whale optimization algorithm to optimize the hyperparameters of the bidirectional long short-term memory network model;Use the optimized BiLSTM model to predict internet of things perception data.The application can efficiently predict the future trend of internet of things perception data, and is suitable for traffic monitoring, intelligent environment perception and other scenes, and provides reliable decision support for internet of things applications.

Description

Technical Field

[0001] This invention belongs to the field of Internet of Things (IoT) technology and relates to a method for predicting IoT sensing data based on deep learning. Background Technology

[0002] With the rapid development of the Internet of Things (IoT), massive amounts of sensor data are generated and used in various application scenarios, such as smart homes, smart cities, industrial automation, and environmental monitoring. Timely and accurate prediction of this data is crucial for achieving more efficient resource management, early warning systems, and improved system response speed. However, due to the nonlinear, non-stationary, and complex nature of sensor data, traditional prediction methods struggle to effectively process and predict complex sensor data patterns. Furthermore, sensor data is often affected by noise and missing data, further limiting the accuracy of data predictions.

[0003] In recent years, deep learning technology has demonstrated significant advantages in processing large-scale, complex data. Through automatic feature extraction and hierarchical representation, deep learning algorithms can capture the deep structural features of data, making them suitable for processing multi-dimensional and time-series data. Summary of the Invention

[0004] In view of this, the purpose of this invention is to provide a deep learning-based method for predicting IoT sensing data. This method has advantages such as high efficiency, accuracy and robustness, in order to address the above challenges and provide better predictive support for IoT applications.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A deep learning-based method for predicting IoT sensing data, comprising the following steps:

[0007] S1. Acquire IoT sensing data and preprocess it using linear interpolation and CEEMDAN-Boxplot mutation point detection methods;

[0008] S2. Based on the seasonality of IoT data, STL is used to decompose time series data and extract periodicity, trend and residual features.

[0009] S3. Improve the whale algorithm by using nonlinear convergence factor, adaptive weight strategy and random difference mutation strategy respectively;

[0010] S4. Establish a bidirectional long short-term memory network model and use an improved whale optimization algorithm for hyperparameter optimization;

[0011] S5. Using periodicity, trend and residual features as input, the bidirectional long short-term memory network model optimized by the improved whale optimization algorithm is used to predict the features, and the prediction results are linearly added to obtain the final prediction result.

[0012] Furthermore, in step S1, linear interpolation is used to complete the collected IoT sensing data, wherein the linear interpolation method is expressed as:

[0013]

[0014] Among them, y k For the missing data y k+n For y k The next adjacent non-empty data, and y k Difference of n data points; y k-m For y k The previously adjacent non-empty data, and y k They differ by m data points;

[0015] The CEEMDAN-Boxplot mutation point detection method is used to remove noise from the data. The process is as follows: CEEMDAN decomposition is performed on the IoT sensing data sample x(t) to obtain an IMF signal and a residual component signal. The IMF signal frequency is distributed from high to low. The high-frequency part of the IMF signal with an amplitude greater than a threshold is removed by Boxplot, and then the data sample is reconstructed to obtain the reconstructed data sample.

[0016]

[0017] Where M represents the total number of IMF signals obtained from CEEMDAN decomposition and the reconstructed data samples. After removing the first k sorted IMF signals, the value k is calculated as follows:

[0018]

[0019] Where DMSE(n) represents the discrete mean square error of the nth IMF component, and high-frequency and low-frequency components are distinguished by the discrete mean square error criterion, the formula of which is:

[0020]

[0021] Where DMSE(n) represents the discrete mean square error of the nth IMF component; L is the number of original data; x(t) represents the tth data of the original signal; IMF n (t) represents the t-th data point of the n-th IMF component; M is the number of IMF components.

[0022] Furthermore, in step S2, the reconstructed data sample data Seasonal trend decomposition of time series data Y is performed through multiple iterations using nested double loops. The inner loop obtains the decomposed trend and period, while the outer loop updates a weight to ensure the robustness of the algorithm. The process is represented as follows:

[0023] Y = T t +S t +R t

[0024] Where Y represents the IoT sensing data obtained after preprocessing, and T t Indicates the trend term; S t Represents a periodic term; R t This represents the residual term.

[0025] Furthermore, in step S3, the improved whale optimization algorithm IWOA's process is as follows:

[0026] S31. Initialize the whale population, including N randomly generated individual whales, and initialize the parameters, setting the maximum number of iterations t. max And set the initial round t = 1;

[0027] S32. In each iteration, design a nonlinear convergence factor a to update the values ​​of coefficient vectors A and C, and then control the search method of the whale algorithm according to |A|.

[0028] S33. Adopt an adaptive weighting strategy to update individual positions differently based on different individuals and different prey positions;

[0029] S34. Update the position in round t = t + 1, and then update the individual position again through a random difference mutation strategy, taking the optimal position before and after the change;

[0030] S35. Judgment: When t≥t max When t < t, output the optimal solution; when t < t max At that time, calculate the fitness of each individual, record the position of the best individual, and return to step S32.

[0031] Furthermore, in step S32, the nonlinear convergence factor is expressed as:

[0032]

[0033] Where r1 and r2 are random numbers in the range [0, 1], and a is the improved nonlinear convergence factor;

[0034] The update formulas for coefficient vectors A and C are expressed as follows:

[0035] A = 2a·r1-a

[0036] C = 2r²

[0037] Here, A is used to control the whale's position update in the search space, and C is used to guide the attraction relationship between the whale and the current optimal solution. When |A|≤1, local search is used, and when |A|>1, global search is used.

[0038] Furthermore, in step S33, firstly, a dynamic decay factor ω(t) is added, which gradually decreases as the iteration number t increases. Its expression is:

[0039]

[0040] Where ω(t) is the nonlinear inertia weight and α(t) is the correction coefficient;

[0041] If a random value p is generated in the range [0, 1], and p < 0.5, then |A| < 1.

[0042] X(t+1)=ω(t)·X p (t)-A·D

[0043] Where A is a coefficient vector used to control the whale's position update in the search space; D is a coefficient vector representing the distance between the current whale and the optimal solution; X p (t) is the prey position vector, and X(t+1) is the individual position vector updated at each stage;

[0044] If p < 0.5, |A| ≥ 1, then:

[0045] D = |C·X rand (t)-X(t)|,X(t+1)=X rand (t)-A·D

[0046] Among them, X rand Let be the position vector of a whale individual randomly selected from the current group;

[0047] If p ≥ 0.5, then:

[0048] X(t+1)=ω(t)·X p (t)+D·e bl ·cos(2πr)

[0049] Among them, e bl is a random constant, b is a constant used to limit the shape of the logarithmic spiral, and l is a random number in the range [-1, 1].

[0050] Furthermore, in step S34, the process of updating the individual location again using the random difference mutation strategy is represented as follows:

[0051] X(t+1)=l1×(X p (t)-X(t))+l2×(X * (txt))

[0052]

[0053] Among them, X * (t) represents a random individual in the population, f best It is the optimal fitness of the group, f avg is the average fitness of the population, l1 and l2 represent the weight coefficients respectively, l1 weight dominates the local search, l2 weight dominates the global guidance, and ∈ is the minimum value to prevent division by zero.

[0054] Furthermore, in step S4, the bidirectional long short-term memory network model includes a forward long short-term memory network layer and a backward long short-term memory network layer, wherein the input time series is X = {x1, x2, ..., x...} t}, where t is the length of the time series;

[0055] Forward LSTM is:

[0056]

[0057] in, This represents the hidden vector of the feedforward LSTM at time step t;

[0058] Backward LSTM is:

[0059]

[0060] in, This is the hidden vector of the backward LSTM at time step t;

[0061] Final output:

[0062]

[0063] Among them, W forward W is the weight matrix from the forward hidden state to the output. backward Let b be the weight matrix from the backward hidden state to the output. y y is the bias vector of the output layer. t The final output at time step t.

[0064] Furthermore, in step S4, the improved Whale Optimization Algorithm (IWOA) is used to optimize the hyperparameters of the bidirectional long short-term memory network model. The hyperparameters include: the number of hidden units H, the maximum training period E, and the learning rate η. The values ​​of H, E, and η are dynamically adjusted by IWOA to optimize the hyperparameters with the goal of achieving the optimal balance between prediction accuracy and computational efficiency.

[0065] Furthermore, in step S5, the optimized bidirectional long short-term memory network model will incorporate the periodic term T. t Trend item S t Residual term R t As input, each set of predictions is used to make corresponding predictions. Finally, the three sets of prediction results are linearly added together to obtain the final prediction result:

[0066]

[0067] in, This represents the result of the final combined model.

[0068] The beneficial effects of this invention are as follows:

[0069] 1. This invention proposes a combined prediction method based on STL-IWOA-BiLSTM, which better captures the seasonal and trend characteristics of IoT sensing data.

[0070] 2. This invention improves the optimization performance of the combined model based on WOA by using a nonlinear convergence factor, an adaptive weighting strategy, and a random difference mutation strategy. IWOA has a higher convergence speed and global search capability, enabling the model to find more suitable data prediction parameters, avoiding manual parameter tuning, and improving the accuracy and stability of the model.

[0071] 3. The BiLSTM model combining STL decomposition and IWOA optimization exhibits stronger robustness when dealing with complex, nonlinear, and noisy data, and maintains high prediction stability under conditions of missing data and anomalies. This combined model can effectively address noise interference and missing data issues in different scenarios, thereby improving the reliability of predictions.

[0072] 4. The STL-IWOA-BiLSTM combined model of the present invention has good versatility and can adapt to the data prediction needs of various Internet of Things scenarios, such as environmental monitoring, energy management and intelligent manufacturing, providing high-precision data prediction support for different applications.

[0073] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0074] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein:

[0075] Figure 1 A flowchart of the method provided by the present invention;

[0076] Figure 2 This is a flowchart of the improved whale algorithm IWOA of this invention. Detailed Implementation

[0077] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0078] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0079] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0080] Please see Figures 1-2This is a deep learning-based method for predicting IoT sensing data.

[0081] Example

[0082] This embodiment proposes a specific implementation process for an IoT sensing data prediction method based on deep learning, such as... Figure 1 As shown, it includes the following steps:

[0083] Step S1: Acquire IoT sensing data and preprocess it using linear interpolation and CEEMDAN-Boxplot mutation detection methods;

[0084] Because IoT sensing data is prone to data loss and anomalies during the collection process, linear interpolation is used to complete the data and CEEMDAN-Boxplot mutation point detection is used to remove the influence of noise on the experimental results in order to better train the data.

[0085] The linear interpolation method is as follows:

[0086]

[0087] Among them, y k For the missing data y k+n For y k The next adjacent non-empty data, and y k Difference of n data points; y k-m For y k The previously adjacent non-empty data, and y k The difference is m data points.

[0088] The CEEMDAN-Boxplot mutation point detection method is as follows: The IoT sensing data sample x(t) is decomposed using CEEMDAN to obtain an IMF signal and a residual component signal. The IMF signal frequency is distributed from high to low. The high-frequency components of the IMF with excessively large amplitudes are removed using Boxplot before reconstruction. High and low frequencies are distinguished using the discrete mean square error criterion, the formula of which is:

[0089]

[0090] Where: DMSE(n) represents the discrete mean square error of the nth IMF component; L is the number of original data; x(t) represents the tth data of the original signal; IMF n (t) represents the t-th data point of the n-th IMF component; M is the number of IMF components.

[0091] DMSE represents the reconstruction error between the IMF component and the original signal. IMF components with smaller DMSE values ​​correspond to the main components of the original signal, while those with larger DMSE values ​​correspond to noise or subtle details. A value k is calculated using the following formula, such that the portion after the k-th IMF component is a low-frequency signal that retains valuable signal information, while the high-frequency components mainly contain noise and abrupt changes. The expression for k is:

[0092]

[0093] Reconstructed data samples of signals after removing noise and abrupt changes Represented as:

[0094]

[0095] Step S2: Use Seasonal and Trend Decomposition using Loess (STL) on the preprocessed time series data. Decompose;

[0096] IoT sensing data typically exhibits seasonality and trends. STL decomposition can break down the data into periodic, trend, and residual components. The periodic component is relatively stable and constitutes a large proportion of the data, while the trend component is unstable; some data show a clear trend, while others do not. Therefore, directly training and predicting from the initial time-series data can weaken the trend component during model learning, potentially leading to predictions that deviate from the original trend.

[0097] The STL decomposition formula is as follows:

[0098] Y = T t +S t +R t

[0099] Where Y represents the IoT sensing data obtained after preprocessing, and T t Indicates the trend term; S t Represents a periodic term; R t This represents the residual term. The decomposition is achieved through multiple iterations using nested double loops. The inner loop obtains the trend and periodicity of the decomposed components, while the outer loop updates a weight to ensure the robustness of the algorithm.

[0100] Step S3: Improve the Improved Whale Optimization Algorithm (IWOA) using a nonlinear convergence factor, an adaptive weighting strategy, and a random difference mutation strategy, respectively; for example... Figure 2The flowchart of the improved whale algorithm IWOA of the present invention shown includes the following steps:

[0101] Step S3.1: Form a whale pod, consisting of N randomly generated individual whales, and initialize the parameters, setting the maximum number of iterations t. max Let t = 1;

[0102] Step S3.2: Update the values ​​of A and C according to the proposed nonlinear convergence factor. A and C are coefficient vectors. A controls the position update of the whale in the search space, and C guides the attraction relationship between the whale and the current optimal solution. In WOA, the value of |A| determines whether to use global random search or local search. In the early stage of iteration, more global search is needed, while in the later stage, more local search is needed. In the original WOA, the value of |A| is controlled by a value that decreases linearly from 2 to 0, which cannot effectively control the global and local search. Therefore, this invention proposes a nonlinear convergence factor that decreases nonlinearly from 2 to 0, as shown in the equation:

[0103]

[0104] A = 2a·r1-a

[0105] C = 2r²

[0106] Where r1 and r2 are random numbers in the range [0, 1], and a is the improved nonlinear convergence factor, which controls the magnitude of |A|. When |A| ≤ 1, local search is used; when |A| > 1, global search is used. The coefficient 'a' has a slower convergence speed in the early stage, which can expand the search range and improve the algorithm's global search capability. In the later stage, the convergence speed is faster, which can improve the solution efficiency, making the algorithm's local mining capability stronger, thus improving the algorithm's search capability while accelerating the convergence speed.

[0107] Step S3.3: Randomly generate a value p within the range [0, 1]. Update the position differently depending on the individual and the location of the prey, using an adaptive weighting strategy. Add a dynamic decay factor ω(t), which gradually decreases as the number of iterations t increases. Its expression is:

[0108]

[0109] Where ω(t) is the nonlinear inertia weight, t max α(t) represents the maximum number of iterations, and α(t) is the correction coefficient.

[0110] The expression for the spiral update phase is then:

[0111] When p < 0.5 and |A| < 1:

[0112] X(t+1)=ω(t)·X p (t)-A·D

[0113] Where A is a coefficient vector used to control the whale's position update in the search space, D is a coefficient vector representing the distance between the current whale and the optimal solution, and X... p (t) is the prey position vector, and X(t+1) is the individual position vector updated at each stage.

[0114] When p < 0.5 and |A| ≥ 1:

[0115] D = |C·X rand (t)-X(t)|,X(t+1)=X rand (t)-A·D

[0116] Where: X rand Let be the position vector of a whale individual randomly selected from the current group.

[0117] When p ≥ 0.5:

[0118] X(t+1)=ω(t)·X p (t)+D·e bl ·cos(2πr)

[0119] Among them, e bl is a random constant, b is a constant used to limit the shape of the logarithmic spiral, and l is a random number in the range [-1, 1].

[0120] Step S3.4: Individuals are updated again using a random difference mutation strategy, and their optimal positions before and after the change are taken. Let t = t + 1, which speeds up population convergence and effectively prevents the population from getting trapped in local optima. Its expression is:

[0121] X(t+1)=l1×(X p (t)-X(t))+l2×(X * (txt))

[0122]

[0123] Among them, X * (t) represents a random individual in the population, f best It is the optimal fitness of the group, f avg is the average fitness of the population, l1 and l2 represent the weight coefficients respectively, l1 weight dominates the local search, l2 weight dominates the global guidance, and ∈ is the minimum value to prevent division by zero.

[0124] Step S3.5: Determine if t≥t max When t < t, output the optimal solution; when t < t maxAt that time, the fitness of each individual is calculated and the optimal individual position is recorded.

[0125] Step S4: Establish a bidirectional long short-term memory (BiLSTM) network model, which is used to predict IoT sensing data based on the extracted periodic, trend, and residual features.

[0126] The bidirectional long short-term memory network model includes a forward long short-term memory network layer and a backward long short-term memory network layer, wherein,

[0127] The input time series is X = {x1, x2, ..., x...} t}, where t is the length of the time series.

[0128] Forward LSTM is:

[0129]

[0130] in, This is the hidden vector of the feedforward LSTM at time step t.

[0131] Backward LSTM is:

[0132]

[0133] in, This is the hidden vector of the backward LSTM at time step t.

[0134] Final output:

[0135]

[0136] Among them: W forward W is the weight matrix from the forward hidden state to the output. backward Let b be the weight matrix from the backward hidden state to the output. y y is the bias vector of the output layer. t The final output at time step t.

[0137] The IWOA algorithm in step S3 is used to optimize BiLSTM. The IWOA algorithm is used to optimize the hyperparameters of BiLSTM, such as the number of hidden units, the maximum training period, and the learning rate, through grid search.

[0138] S5: The BiLSTM model optimized by the IWOA algorithm takes the periodic term, trend term, and residual term obtained from the STL decomposition in step S2 as inputs and makes corresponding predictions. Finally, the three sets of prediction results are linearly added to obtain the final prediction result.

[0139]

[0140] in, This represents the result of the final combined model.

[0141] This embodiment also provides the following evaluation metrics for the final STL-IWOA-BiLSTM combined prediction model:

[0142]

[0143]

[0144] Where yi is the true value of the predicted object. The predicted value after inverse normalization. The actual mean of the predicted object is n, where n is the amount of data.

[0145] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for predicting IoT sensing data based on deep learning, characterized in that: The method includes the following steps: S1. Acquire IoT sensing data and preprocess it using linear interpolation and CEEMDAN-Boxplot mutation point detection methods; S2. Based on the seasonality of IoT data, STL is used to decompose time series data and extract periodicity, trend and residual features. S3. Improve the whale algorithm using a nonlinear convergence factor, an adaptive weighting strategy, and a random difference mutation strategy, respectively. In step S3, the improved whale optimization algorithm IWOA's flow is as follows: S31. Initialize the whale population, including N randomly generated individual whales, and initialize the parameters, setting the maximum number of iterations. And set the initial round. ; S32. In each iteration, design a nonlinear convergence factor. Update coefficient vector and The value, and then according to Controlling the search method performed by the whale algorithm; S33. Adopt an adaptive weighting strategy to update individual positions differently based on different individuals and different prey positions; S34, Update Round Then, the individual position is updated again using a random difference mutation strategy, and the optimal position before and after the change is taken. S35. Judgment: When When, output the optimal solution; when At that time, calculate the fitness of each individual, record the position of the best individual, and return to step S32; S4. Establish a bidirectional long short-term memory network model and perform hyperparameter optimization using an improved whale optimization algorithm. In step S4, the bidirectional long short-term memory network model includes a forward long short-term memory network layer and a backward long short-term memory network layer, wherein the input time series is... , The length of the time series; Forward LSTM is: in, For the forward LSTM at time step The hidden vector; Backward LSTM is: in, For backward LSTM at time step The hidden vector; Final output: in, This is the weight matrix from the forward hidden state to the output. The weight matrix from the backward hidden state to the output. This is the bias vector for the output layer. Time step The final output; The improved Whale Optimization Algorithm (IWOA) is used to optimize the hyperparameters of the bidirectional Long Short-Term Memory (LSTM) network model. The hyperparameters include the number of hidden units. Maximum training cycle and learning rate Dynamically adjust via IWOA , , The hyperparameters are optimized with the goal of achieving the optimal balance between prediction accuracy and computational efficiency. S5. Using periodicity, trend and residual features as input, the bidirectional long short-term memory network model optimized by the improved whale optimization algorithm is used to predict the features, and the prediction results are linearly added to obtain the final prediction result.

2. The IoT sensing data prediction method based on deep learning according to claim 1, characterized in that: In step S1, linear interpolation is used to complete the collected IoT sensing data. The linear interpolation method is expressed as follows: in, For missing data for The next adjacent non-empty data, and difference One data point; for The previously adjacent non-empty data, and difference One data point; The CEEMDAN-Boxplot mutation point detection method was used to remove noise from the data. The process was as follows: For IoT sensing data samples... CEEMDAN decomposition yields an IMF signal and a residual component signal. The IMF signal frequency is distributed from high to low. Boxplot is used to remove IMF signals with high-frequency amplitudes exceeding a threshold before reconstruction to obtain the reconstructed data sample. : in, This represents the total number of IMF signals obtained from CEEMDAN decomposition, and the reconstructed data samples. Remove the first sorted part One IMF signal, value The calculation method is as follows: in, Indicates the first The discrete mean square error of each IMF component, high frequency and low frequency are distinguished by the discrete mean square error criterion, the formula of which is: in, Indicates the first The discrete mean square error of each IMF component; The number of original data items; The first part represents the original signal. One data point; Indicates the first The first IMF component One data point; This represents the number of IMF components.

3. The IoT sensing data prediction method based on deep learning according to claim 2, characterized in that: In step S2, the reconstructed data sample data As time series data Seasonal trend decomposition is performed through multiple iterations using nested double loops. The inner loop obtains the decomposed trend and period, while the outer loop updates a weight used to ensure the robustness of the algorithm. The process is represented as follows: in, This represents the IoT sensing data obtained after preprocessing. Indicates the trend; Indicates a periodic term; This represents the residual term.

4. The IoT sensing data prediction method based on deep learning according to claim 1, characterized in that: In step S32, the nonlinear convergence factor is expressed as: in, , Random numbers in the range [0, 1] This refers to the improved nonlinear convergence factor. coefficient vector and The update formula is expressed as: in, Used to control the whale's position updates in the search space. Used to guide the attraction relationship between the whale and the current optimal solution, when When, a local search is used, when When this is the case, a global search is used.

5. The IoT sensing data prediction method based on deep learning according to claim 4, characterized in that: In step S33, firstly, a dynamic attenuation factor is added. As the number of iterations increases As it increases, it gradually decreases, and its expression is: in, For nonlinear inertia weights, This is a correction factor; Randomly generate a value within the range [0,1]. ,like , ,have: in, This is a coefficient vector used to control the whale's position update in the search space; The coefficient vector represents the distance between the current whale and the optimal solution; Let be the prey's position vector. The updated individual position vectors at each stage; like , ,have: in, Let be the position vector of a whale individual randomly selected from the current group; like ,have: in, It is a random constant. The constant used to define the shape of the logarithmic spiral, A random number in the range [-1, 1].

6. The IoT sensing data prediction method based on deep learning according to claim 5, characterized in that: In step S34, the process of updating the individual location again using the random difference mutation strategy is represented as follows: , in, A random individual in the population. It is the optimal fitness of the group. It is the average fitness of the population. and These represent the weighting coefficients, Weights dominate local search. Weighting dominates the overall guidance. It is a minimum value to prevent division by zero.

7. The IoT sensing data prediction method based on deep learning according to claim 1, characterized in that: In step S5, the optimized bidirectional long short-term memory network model will include periodic terms. Trend items residuals As input, each set of predictions is used to make corresponding predictions. Finally, the three sets of prediction results are linearly added together to obtain the final prediction result: in, This represents the result of the final combined model.

Images