Green plant growth prediction system and detection method based on multi-parameter environment monitoring

Abstract

The invention discloses a green plant growth prediction system based on multi-parameter environment monitoring and a detection method. The system belongs to the technical field of agricultural environment automatic monitoring and comprises an in-water oxygen content quantification module, a PH value processing module, a soil chemical fertilizer content processing module, a solar radiation value processing module, a green plant growth prediction module, a green plant growth correction module and an in-water oxygen content quantification module. According to the characteristics of green plant environment and oxygen content change in water, green plant planting is detected from a seedling stage, a germination stage, a seedling stage, a flowering and fruiting stage and a vegetative growth stage, and the growth amount of the green plants is predicted; according to the method, the oxygen content in water, the PH value, the chemical fertilizer content in soil and solar radiation required by a green plant growth environment are detected and processed, parameters of an MTFE-PatchTST model are optimized by applying an improved aurora optimization algorithm, and then historical growth data and temperature data of green plants are preliminarily predicted by applying the optimized MTFE-PatchTST model; and finally, performing correction by combining factors obtained by the cross-correlation function module, and accurately predicting the green plant growth amount.

Description

Technical Field

[0001] The present invention belongs to the technical field of automated monitoring of agricultural environments, and relates to a green plant growth prediction system; specifically, it relates to a green plant growth prediction system and a detection method based on multi-parameter environmental monitoring. Background Art

[0002] Green plants, as an indispensable life form in nature, not only add vitality and vigor to the living environment, but also contain rich medicinal value; at the same time, they also have unique pharmacological ingredients; green plants not only enrich the treasure house of traditional Chinese medicine, but also provide rich resources; they have made important contributions to the health cause by regulating human functions, enhancing immunity, preventing and treating diseases, etc.; at different stages of green plant growth, the oxygen content and pH conditions in the water have different effects on the growth of green plants. The judgment of the microclimate environment is a complex process, which directly affects the accuracy of the predicted increase or decrease in green plant growth.

[0003] The oxygen content in water directly affects the respiration and root health of green plants. Adequate dissolved oxygen helps the plant roots absorb nutrients and carry out normal metabolism, promoting the healthy growth of green plants. On the contrary, an oxygen-deficient environment will inhibit plant growth and even cause root rot.

[0004] Solar radiation is the main source of energy for green plants to carry out photosynthesis. Appropriate light can promote the synthesis of chlorophyll, improve the efficiency of photosynthesis, and is beneficial to the growth, flowering and fruiting of green plants. However, excessive light may cause photoinhibition and cause damage to green plants.

[0005] The pH value determines the acidity and alkalinity of the water, which in turn affects the absorption and utilization of nutrients by green plants. Different green plants have different adaptability to pH values, but most green plants prefer a neutral or slightly acidic soil environment. A suitable pH value helps plant roots absorb necessary minerals and trace elements and promote growth.

[0006] The fertilizer content is directly related to the nutrient supply of green plants; the appropriate application of fertilizers can provide the nitrogen, phosphorus, potassium and other nutrients required by plants, and promote plant growth and development; however, excessive application of fertilizers can lead to soil salinization, nutrient imbalance and other problems, which will have an adverse effect on green plants; therefore, reasonable control of fertilizer usage is the key to maintaining the healthy growth of green plants.

[0007] In the existing green plant yield prediction system, there is no system that subdivides the predictions into different growth cycles and growth stages. The prediction accuracy is not high, and the prediction of yield under different water oxygen content conditions in different growth stages has not yet been involved. Summary of the invention

[0008] In view of the above problems, the purpose of the present invention is to propose a detection method for a green plant growth prediction system based on multi-parameter environmental monitoring, which detects and processes the oxygen content in the water required for the growth environment of green plants in different growth periods, and then corrects the prediction results to accurately predict the growth of green plants.

[0009] The technical solution of the present invention is: a green plant growth prediction system based on multi-parameter environmental monitoring described in the present invention comprises a solar radiation data processing module, a pH value data processing module, a fertilizer content data processing module, a water oxygen content quantification module, a cross-correlation function module, a green plant growth prediction module and a green plant growth correction module which are interconnected;

[0010] The solar radiation data processing module includes a solar radiation detection module, a solar radiation static data fusion module and a sliding filter solar radiation data processing device;

[0011] The pH value data processing module includes a pH value detection module, a pH value static data fusion module and a sliding filter pH value data processing device;

[0012] The fertilizer content data processing module includes a fertilizer content detection module, a fertilizer content static data fusion module and a sliding filter fertilizer content data processing device;

[0013] The water oxygen content quantification module includes five water oxygen content quantification modules for seedling stage, germination stage, seedling stage, flowering and fruiting stage and vegetative growth stage;

[0014] The cross-correlation function module is used to measure the similarity between the growth of green plants and solar radiation, fertilizer content in water, oxygen content in water, and pH value of two signals at different time lags;

[0015] The green plant growth prediction module is an MTFE-PatchTST prediction model;

[0016] The green plant growth correction module generates prediction errors after the optimized MTFE-PatchTST preliminary prediction, calculates the error sequence, and then uses the optimized MTFE-PatchTST model to correct the error data again;

[0017] For the solar radiation data processing module, the solar radiation value data of the green plant growth environment is collected through multiple sensors in the solar radiation data processing module, and then the collected pH value is fused through the multi-point solar radiation value static data fusion module to output the solar radiation value after data fusion, and then the data abnormal value is processed through the sliding filter solar radiation value data processing device to output the final solar radiation value;

[0018] For the pH value data processing module, the pH value data of the green plant growth environment is collected through multiple sensors in the pH value data processing module, and then the collected pH values ​​are fused through the multi-point pH value static data fusion module to output the fused pH value, and then the data abnormal values ​​are processed through the sliding filter pH value data processing device to output the final pH value;

[0019] For the fertilizer content data processing module, the fertilizer content data of the green plant growth environment is collected through multiple sensors in the fertilizer content data processing module, and then the collected fertilizer content is fused through the multi-point fertilizer content static data fusion module to output the fertilizer content after data fusion, and then the data abnormal values ​​are processed through the sliding filter fertilizer content data processing device to output the final fertilizer content;

[0020] For the water oxygen content quantification module, the fused oxygen content data is quantified by K-means clustering according to 5 different growth stages to obtain the water oxygen content data in different growth stages;

[0021] The green plant growth prediction module collects and analyzes the historical growth data of green plants and the corresponding historical temperature data, then uses the improved Aurora optimization algorithm to optimize the MTFE-PatchTST scaling factor parameters, and finally uses the optimized MTFE-PatchTST prediction model to perform complex calculations and output prediction results.

[0022] Furthermore, a detection method of a green plant growth prediction system based on multi-parameter environmental monitoring, the implementation process of the water oxygen content quantification module based on K-means clustering quantification is as follows:

[0023] Step (1): randomly select g objects from a oxygen content fusion data as the initial clustering centers;

[0024] Step (2): Calculate the distance from each data point to the cluster center and assign them to the cluster closest to it one by one. The distance expression formula is as follows:

[0025]

[0026] In the formula, y represents the oxygen content fusion data object, E i represents the i-th cluster center of the oxygen content fusion data cluster, p represents the number of growth periods; y j , E ij Represents y and E i The j-th attribute value of ;

[0027] Step (3): After all growth periods are assigned, the mean of all growth periods in each cluster is used as the new cluster center of the category, and the distance is recalculated so that the total square error sum of each sample point to the nearest cluster center is as small as possible. The calculation formula of the total square error sum is as follows;

[0028]

[0029] In the formula, the size of the total error square sum indicates the quality of the clustering result, and h represents the number of clusters;

[0030] Step (4): Compare with the H cluster centers calculated last time. If the cluster centers have changed, go to step (2); otherwise, go to step (5).

[0031] Step (5): When the cluster center no longer changes, stop and output the cluster quantization results.

[0032] Furthermore, the implementation process of the solar radiation data processing module, the pH value data processing module and the fertilizer content data processing module is as follows:

[0033] Step (1): Use sensors to collect pH value, solar radiation, and fertilizer data samples in water;

[0034] Step (2): using static fusion technology to process the sensor's measurement data to obtain processed optimized data;

[0035] Step (3): Use a sliding filter to process the optimized data to obtain the final data;

[0036] Define the final data of each pH sensor, recorded as: y[a] θ1 ; where θ1 is the number of the pH sensor, x[a] θ1 is the input of the pH sliding filter, and a is the number of pH sensors;

[0037] Define the final data of each solar radiation sensor as: y[b] θ2 ; where θ2 is the number of the pH sensor, x[b] θ2 is the input of the solar radiation sliding filter, b is the number of solar radiation sensors;

[0038] Define the final data of each fertilizer data sensor, recorded as: y[c] θ3 ; where θ3 is the number of the fertilizer data sensor, x[c] θ3 is the input of the fertilizer data sliding filter, c is the number of fertilizer sensors;

[0039] The formula of its sliding filter is as follows:

[0040]

[0041] Where A is the size of the pH sliding window, B is the size of the solar radiation sliding window, C is the size of the fertilizer sliding window, and k is a positive integer.

[0042] Furthermore, the algorithm improvement implementation steps of the green plant growth prediction module are as follows:

[0043] Step (1): In the initialization phase, the iterative process of the Aurora optimization algorithm begins with an initial population generated by pseudo-random numbers; this population is represented as a matrix whose dimension is the number of candidate solutions in the population multiplied by the scalable dimension of the solution space, and its formula is as follows:

[0044]

[0045] In the Aurora Optimization Algorithm, the search agent simulates the process of high-energy charged particles moving along magnetic field lines toward the Earth's poles in the solution space. The boundaries of the agent are defined by upper and lower bounds, while R represents a random number sequence between 0 and 1;

[0046] Step (2): During the rotation phase, charged particles approaching the Earth interact with the Earth's magnetic field, causing them to rotate along the magnetic field lines; this phenomenon is described by the formula for the Lorentz force:

[0047]

[0048] In the formula, CJ represents the integral constant, Ec represents the charge of the charged particle, Zl represents the mass of the particle, and Qd represents the strength of the geomagnetic field. These four parameters are all fixed values. Therefore, it is assumed that the values ​​of CJ, Ec, and Qd are all equal to 1, and the value of Zl is set to 100. The damping factor Df is randomly selected in the range of 1 to 1.5, and the time is simulated within the strategy using the algorithm's fitness evaluation process.

[0049] Step (3): In the rotational motion stage, in order to explore the local solution space, find the local optimum or optimize the local structure of the current solution; introduce the aurora ellipse walking strategy to improve local development and fine-tune to obtain the best solution or local optimum;

[0050] This strategy focuses on global search capabilities, and particles take larger steps to explore in the solution space; this reflects the global search phase of the algorithm, in which the solution space is extensively searched with larger steps in order to locate the global optimal solution;

[0051] The improved PLO algorithm has the following formula:

[0052] JZ new(i,j)=X(i,j)+r2×(W1×v(t)+W2×Ao)

[0053] Ao=Levy(d)×(JZ avg (j)-JZ(i,j))+LB+r1×(UB-LB) / 2

[0054]

[0055] In the formula, JZ new (i,j) represents the position of the high-energy particle after the update, r2 represents the interference of uncontrollable environment and other factors on the particle, and takes the value of [0,1];

[0056] Step (4): During the particle collision phase, the chaotic collision between particles enables the PLO to leave the local optimum. When these particles enter the atmosphere and merge with the aurora in the ellipse, the formula is as follows:

[0057] JZ new (i,j)=JZ(i,j)+sin(r3×π)×(JZ(i,j)-JZ(a,j)),r4<Kandr5<0.05

[0058]

[0059] Where JZ(a,j) refers to a random particle in the particle swarm; as the algorithm progresses, the frequency of collisions between particles gradually increases, and this process is regulated by the collision probability K; the value range of r3 and r4 is [0,1].

[0060] Furthermore, the process of optimizing the scaling factor D of the MTFE-PatchTST model using the improved Aurora optimization algorithm is as follows:

[0061] Initialize the relevant parameters of the improved Aurora optimization algorithm, including the initial particle position, the upper and lower limits of the search space, the maximum number of iterations, and the current number of iterations;

[0062] The training set of green plant growth data is input into the MTFE-PatchTST model to optimize the scaling factor D of the model;

[0063] The new particle position is obtained in the particle collision stage of the improved Aurora optimization algorithm, and the fitness value of each particle is calculated. When the preset termination condition of the model is met, the optimal particle position is determined, which is the optimal value of the scaling factor D of the MTFE-PatchTST model.

[0064] The optimized MTFE-PatchTST model predicts the data after K-means aggregation and quantization to obtain preliminary green plant growth prediction results.

[0065] Furthermore, the implementation steps of the green plant growth prediction module are as follows:

[0066] A multivariate time series feature extraction module MTFE consisting of an encoder and a decoder is introduced into the PatchTST model as an extension and improvement of the basic model to form the MTFE-PatchTST model;

[0067] This extended improvement enhances the model's data processing capabilities, enabling it to extract features from complex multivariate time series, thereby improving the prediction accuracy of green plant growth; the specific process is as follows:

[0068] Step (1): establish a multivariate time series sample set of green plant growth with a lookback window L, where each sample in the sample set contains multiple variables, and the above variables together describe the growth status of the green plants;

[0069] Each sample can be represented as a multivariate time series, where the length of each variable sequence is L, and the green plant growth sample is set as:

[0070] t={t1,t2,…,t L};

[0071] Step (2): The variable sequence of length L starting from index 1 is expressed as Where k is a positive integer, l specifically refers to the position of the variable in L, and M refers to the total number of variables in the variable sequence; input t = {t1, t2, …, t L} is separated into univariate sequences t (l) ∈R 1×L , R refers to a real number sequence with 1 row and L columns;

[0072] Among them, each sequence enters the Transformer network according to the channel-independence setting;

[0073] Step (3): Input the patch-separated historical growth data of green plants into the encoder of the MTFE module and evaluate it through the self-attention mechanism. The evaluation formula is as follows:

[0074]

[0075] In the formula, Q is Query, K is Key, V is Value, and D is the scaling factor;

[0076] Step (4): The output of the encoder is then fed into the MTFE Decoder, which uses a self-attention mechanism to further process the features.

[0077] Finally, a fully connected layer with a linear head is used to obtain the predicted result, that is, the amount of green plant growth at a certain point in the future:

[0078]

[0079] Furthermore, the implementation process of the cross-correlation function module is as follows:

[0080] Step (1): Input discrete signal acquisition data value X m (p),

[0081] In the formula, m = 1, 2, 3, 4 represent the input oxygen content, pH value, solar radiation and fertilizer data respectively, and p represents the corresponding time of the mth group of data collection;

[0082] Step (2): Input historical data value Y n (q),

[0083] In the formula, n=1, 2 represent the input green plant growth and water temperature historical data respectively.

[0084] q represents the corresponding time of the nth group of historical data;

[0085] Step (3): Zero-mean the signal to eliminate the influence of the DC component. The expression is as follows:

[0086]

[0087] In the formula, represents the mean value of the input signal X in the mth group of data, Represents the mean value of the input signal Y in the nth group of data;

[0088] Step (4): Calculate the correlation coefficient between the input arrays and determine the input signal X m (p) and input signal Y n (q), select the data with greater correlation as the input of the MTFE-PatchTST prediction module, and the calculation formula is as follows:

[0089] R X'Y' (θ) = ∑[X' m (p)·Y' n (q-θ)]

[0090] In the formula, R X'Y' (θ) represents the cross-correlation function value at the position with lag order θ, R X'Y' The larger the value of (θ), the stronger the cross-correlation.

[0091] Furthermore, the implementation process of the green plant growth correction module is as follows:

[0092] First, use each module to collect data of green plants at different growth stages, and use K-means to aggregate and organize the data;

[0093] Then, the optimized MTFE-PatchTST is used to make a preliminary prediction of the aggregated data; the actual green plant growth data is subtracted from the preliminary prediction results to obtain the error time series; the expression of this process is as follows:

[0094] Error=Ovalue-Ivalue

[0095] In the formula, Error represents the error value of the green plant growth data, Ovalue represents the measured value, and Ivalue represents the initial predicted value;

[0096] The error time series is reorganized into input, and then the optimized MTFE-PatchTST model is used to predict the Error using the actual green plant growth data to obtain the error prediction result IError of the green plant growth;

[0097] Finally, the initial prediction result and the error prediction result are added together to get the final prediction result, which is as follows:

[0098] Fvalue=Ivalue+IError

[0099] Where Fvalue represents the predicted value of the final green plant growth.

[0100] The beneficial effects of the present invention are as follows: 1. In view of the randomness of problems such as sensor interference and abnormal measurement values ​​in the process of measuring the growth environment parameters of green plants, the present invention subdivides green plants into 5 different growth periods, and processes the growth environment parameters measured by the sensor in these 5 growth periods through a sliding filter module, thereby effectively dealing with the uncertain factors occurring in the growth environment of green plants; improving the objectivity and credibility of the detection parameters of the sensor values ​​for detecting oxygen content, pH value, light, and fertilizer parameters in the growth environment of green plants; 2. The present invention adopts the MTFE-PatchTST prediction model to detect the growth environment of green plants according to the detected green plant planting environment. The temperature data of the environment and the historical growth data are used to predict the growth situation at different temperatures, and the improved Aurora optimization algorithm is used to optimize the MTFE-PatchTST scaling factor parameters, which can more accurately grasp the changing trend of green plant growth and facilitate the impact of temperature on growth, so as to improve the oxygen content, pH value, light, and fertilizer production management of the plantation environment; 3. The K-means oxygen content quantification module adopted in the present invention realizes the quantification of the oxygen content data of the green plant growth environment being detected into 5 growth stage data of the green plants, so as to better control the oxygen content in the water during the entire growth process at different stages. BRIEF DESCRIPTION OF THE DRAWINGS

[0101] Figure 1 It is a schematic diagram of the structure of the present invention;

[0102] Figure 2 is a flow chart of a data processing module for fertilizer content in soil according to an embodiment of the present invention;

[0103] Figure 3 is a flow chart of a solar radiation data processing module and a pH value data processing module in an embodiment of the present invention;

[0104] Figure 4 is a flow chart of a module for quantifying oxygen content in water according to an embodiment of the present invention;

[0105] Figure 5 is a structural diagram of a green plant growth prediction module in an embodiment of the present invention;

[0106] Figure 6 is a flow chart of a green plant growth correction model according to an embodiment of the present invention;

[0107] Figure 7 It is a flow chart of the improved Aurora optimization algorithm in an embodiment of the present invention. DETAILED DESCRIPTION

[0108] The specific technical scheme of the present invention is further described in detail below with reference to specific examples.

[0109] As shown in the figure, this embodiment provides a detection method of a green plant growth prediction system based on multi-parameter environmental monitoring, including a pH value data processing module; a fertilizer content data processing module; a light processing module (solar radiation data processing module); a water oxygen content quantification module, a green plant growth prediction module and a green plant growth correction module;

[0110] The pH value processing module includes an oxygen content detection module and a sliding filter device module (a pH value detection module, a pH value static data fusion module and a sliding filter pH value data processing device), collects pH value data of the green plant growth environment through multiple sensors in the pH value detection module, and then processes the collected pH value data through the sliding filter device module to output the pH value after data fusion; performs fusion processing, outputs the pH value after data fusion, and then processes the data abnormal value through the sliding filter pH value data processing device to output the final pH value;

[0111] The light processing module (solar radiation data processing module) includes an oxygen content detection module and a sliding filter device module (solar radiation detection module, solar radiation static data fusion module and sliding filter solar radiation data processing device). The light value data of the green plant growth environment is collected through multiple sensors in the light value detection module, and then the collected light value data is processed by the sliding filter device module to output the light value after data fusion, and then the data abnormal value is processed by the sliding filter solar radiation value data processing device to output the final solar radiation value;

[0112] The fertilizer content data processing module includes an oxygen content detection module and a sliding filter device module (a fertilizer content detection module, a fertilizer content static data fusion module and a sliding filter fertilizer content data processing device). The fertilizer value data of the green plant growth environment is collected through multiple sensors in the fertilizer value detection module, and then the collected fertilizer value data is processed by the sliding filter device module to output the fertilizer value after data fusion, and then the data abnormal value is processed by the sliding filter fertilizer content data processing device to output the final fertilizer content;

[0113] The oxygen content quantification module in water includes five modules: seedling stage, germination stage, seedling stage, flowering and fruiting stage, and vegetative growth stage. The fused oxygen content data are then quantified by K-means clustering according to the five different growth stages to obtain the oxygen content data in water at different growth stages.

[0114] The green plant growth prediction module is the MTFE-PatchTST prediction model; it includes collecting historical temperature data and historical growth data. This module collects and analyzes the historical growth data of green plants and the corresponding historical temperature data, and then uses the improved Aurora optimization algorithm to optimize the MTFE-PatchTST scaling factor parameters. Finally, the optimized MTFE-PatchTST prediction model performs complex calculations to obtain the initial prediction results. This model is specially optimized for the characteristics of green plant growth and can more accurately capture the dynamic relationship between temperature and green plant growth.

[0115] The cross-correlation function module is used to measure the similarity between the growth of green plants and solar radiation, fertilizer content in water, oxygen content in water, and pH value of two signals at different time lags, select the factors with the greatest influence and calculate the error sequence of the data; it has a wide range of applications in signal processing, time series analysis, and image processing. By measuring the factors most relevant to the growth of green plants, it is used in the following green plant growth correction module to measure the growth of green plants more accurately.

[0116] The green plant growth correction model includes using various modules to collect data from different growth periods of green plants, and using K-means to aggregate and organize the data; then using the optimized MTFE-PatchTST to make a preliminary prediction on the aggregated data; subtracting the actual green plant growth data from the preliminary prediction results to obtain an error time series, reorganizing the input through the obtained error time series, and then using the optimized MTFE-PatchTST model to predict Error using the actual green plant growth data to obtain the error prediction result IError of green plant growth; that is: the green plant growth correction module is a preliminary prediction of the optimized MTFE-PatchTST, which will produce a prediction error. After calculating the error series, the optimized MTFE-PatchTST model is used again to correct the error data; this method can reduce the error impact caused by the initial prediction and improve the accuracy of the prediction.

[0117] Furthermore, a detection method of a green plant growth prediction system based on multi-parameter environmental monitoring and an oxygen content quantification module based on K-means aggregation are implemented as follows:

[0118] L1: Randomly select g objects from a oxygen content fusion data as the initial clustering centers;

[0119] L2: Calculate the distance from each data point to the cluster center and assign them to the cluster closest to it one by one; the distance expression formula is as follows:

[0120]

[0121] Where y is the oxygen content fusion data object, E i is the i-th cluster center of the oxygen content fusion data cluster, p is the number of growth periods; y j , E ij for y and E i The j-th attribute value of ;

[0122] L3: After all growth periods are assigned, the mean of all growth periods in each cluster is used as the new cluster center of the category, and the distance is recalculated so that the total square error sum of each sample point to the nearest cluster center is as small as possible. The calculation formula for the total square error sum is as follows;

[0123]

[0124] In the formula, the size of the total square error (SSE) indicates the quality of the clustering result, and h is the number of clusters;

[0125] L4: Compare with the H cluster centers calculated last time. If the cluster center changes, go to step 2, otherwise go to step 5.

[0126] L5: When the cluster center no longer changes, stop and output the cluster quantization results.

[0127] Furthermore, the solar radiation, pH value and fertilizer content data processing module is implemented as follows:

[0128] S1: Use sensors to collect pH value, solar radiation, and fertilizer data samples in water;

[0129] S2: Use static fusion technology to process the sensor's measurement data to obtain processed optimized data;

[0130] S3: Use sliding filter to process the optimized data to obtain the final data;

[0131] Define the final data of each pH sensor, recorded as: y[a] θ1 ;

[0132] Where θ1 is the number of the pH sensor, x[a] θ1 is the input of pH sliding filter (pH optimization data), a is the number of pH sensors;

[0133] Define the final data of each solar radiation sensor as: y[b] θ2 ;

[0134] Where θ2 is the number of the pH sensor, x[b] θ2 is the input of the solar radiation sliding filter (solar radiation value optimization data), b is the number of solar radiation sensors;

[0135] Define the final data of each fertilizer data sensor, recorded as: y[c] θ3 ;

[0136] Among them, θ3 is the number of the fertilizer data sensor, x[c] θ3 is the input of the fertilizer data sliding filter (fertilizer optimization data), c is the number of fertilizer sensors;

[0137] The formula of its sliding filter is as follows:

[0138]

[0139] Where A is the size of the pH sliding window, B is the size of the solar radiation sliding window, C is the size of the fertilizer sliding window, and k is a positive integer.

[0140] Furthermore, the algorithm improvement implementation steps of the green plant growth prediction module are as follows:

[0141] R1: In the initialization phase, the iterative process of the Aurora Optimization Algorithm (PLO) begins with an initial population generated by pseudo-random numbers; this population is represented as a matrix whose dimension is the number of candidate solutions in the population multiplied by the scalable dimension of the solution space, and the mathematical formula is as follows:

[0142]

[0143] In the PLO algorithm, the search agent simulates the process of high-energy charged particles moving along magnetic field lines toward the Earth's poles in the solution space; the agent's boundaries are defined by an upper bound (UB) and a lower bound (LB), while R represents a sequence of random numbers between 0 and 1;

[0144] R2: During the rotation phase, charged particles approaching the Earth interact with the Earth's magnetic field, causing them to rotate along the magnetic field lines; this phenomenon can be described by the formula for the Lorentz force:

[0145]

[0146] In the formula, CJ represents the integral constant, Ec represents the charge of the charged particle, Zl represents the mass of the particle, and Qd represents the strength of the geomagnetic field. These four parameters are all fixed values. Therefore, it is assumed that the values ​​of CJ, Ec, and Qd are all equal to 1, and the value of Zl is set to 100. The damping factor Df is randomly selected in the range of 1 to 1.5, and the time is simulated within the strategy using the algorithm's fitness evaluation process.

[0147] R3: In the rotational motion stage, in order to explore the local solution space more deeply, find the local optimum or optimize the local structure of the current solution; introduce the Aurora Ellipse Walking Strategy to improve local development and fine-tune to obtain the best solution or local optimum; this strategy focuses on global search capabilities, and particles take larger steps to explore in the solution space, aiming to identify more potential areas;

[0148] This reflects the global search phase of the algorithm, where the solution space is extensively searched with large step sizes in order to locate the global optimal solution;

[0149] The calculation formula of the improved PLO algorithm is as follows:

[0150] JZ new (i,j)=X(i,j)+r2×(W1×v(t)+W2×Ao)

[0151] Ao=Levy(d)×(JZ avg (j)-JZ(i,j))+LB+r1×(UB-LB) / 2

[0152]

[0153] In the formula, JZ new (i,j) represents the position of the high-energy particle after the update, r2 represents the interference of uncontrollable environment and other factors on the particle, and takes the value of [0,1];

[0154] R4: During the particle collision phase, chaotic collisions between particles enable the PLO to leave the local optimum. When these particles enter the atmosphere and merge with the aurora in the ellipse, collisions occur more frequently, causing the shape of the aurora to change continuously; the formula is as follows:

[0155] JZ new (i,j)=JZ(i,j)+sin(r3×π)×(JZ(i,j)-JZ(a,j)),r4<Kandr5<0.05

[0156]

[0157] Where JZ(a,j) refers to a random particle in the particle swarm; as the algorithm progresses, the frequency of collisions between particles gradually increases, and this process is regulated by the collision probability K; the value range of r3 and r4 is [0,1].

[0158] Furthermore, the steps of optimizing the scaling factor D of the MTFE-PatchTST model using the improved Aurora optimization algorithm are as follows:

[0159] Initialize the relevant parameters of the improved Aurora optimization algorithm, including the initial particle position, the upper and lower limits of the search space, the maximum number of iterations, and the current number of iterations;

[0160] The training set of green plant growth data is input into the MTFE-PatchTST model to optimize the scaling factor D of the model;

[0161] The new particle position is obtained in the particle collision stage of the improved Aurora optimization algorithm, and the fitness value of each particle is calculated. When the preset termination condition of the model is met, the optimal particle position is determined, which is the optimal value of the scaling factor D of the MTFE-PatchTST model.

[0162] The optimized MTFE-PatchTST model predicts the data after K-means aggregation and quantization to obtain preliminary green plant growth prediction results.

[0163] Furthermore, the implementation steps of the green plant growth prediction module are as follows:

[0164] A multivariate time series feature extraction module MTFE consisting of an encoder and a decoder is introduced into the PatchTST model as an extension and improvement of the basic model to form the MTFE-PatchTST model; this extension and improvement significantly enhances the model's data processing capabilities, enabling it to more effectively extract features from complex multivariate time series, thereby improving the prediction accuracy of green plant growth. The specific process is as follows:

[0165] V1: Establish a multivariate time series sample set of green plant growth with a lookback window L. Each sample in the sample set contains multiple variables, which together describe the growth status of green plants; each sample can be represented as a multivariate time series, where the length of each variable sequence is L, and the green plant growth sample is set as:

[0166] t={t1,t2,…,t L};

[0167] V2: The variable sequence of length L starting from index 1 is represented as Where k is a positive integer, l specifically refers to the position of the variable in L, and M refers to the total number of variables in the variable sequence; input t = {t1, t2, …, t L} is separated into univariate sequences t (l) ∈R 1×L , R refers to a real number sequence with 1 row and L columns;

[0168] Each sequence enters the Transformer network according to the channel-independence setting;

[0169] V3: The patch-separated historical growth data of green plants is input into the encoder of the MTFE module and evaluated by the self-attention mechanism. The evaluation formula is as follows:

[0170]

[0171] In the formula, Q is Query, K is Key, V is Value, and D is the scaling factor;

[0172] V4: The output of the encoder is then fed into the MTFE Decoder, which also uses the self-attention mechanism to further process the features. Finally, a fully connected layer with a linear head is used to obtain the predicted result, which is the amount of green plant growth at a certain point in the future:

[0173]

[0174] By combining the PatchTST model and the MTFE module, the model can not only capture the long-distance dependencies in the time series data and retain the ability of the PatchTST model to process univariate time series, but also effectively process multivariate time series data, thereby improving the prediction accuracy of green plant growth. This improvement provides a powerful tool for fields such as precision agriculture and environmental monitoring.

[0175] Furthermore, the cross-correlation function module is implemented as follows:

[0176] Q1. Input discrete signal acquisition data value X m (p), m = 1, 2, 3, 4 represent the input oxygen content, pH value, solar radiation and fertilizer data respectively, where p represents the corresponding time of the mth group of data collection;

[0177] Q2. Input historical data value Y n (q), n = 1, 2 represent the input green plant growth and water temperature historical data, where q represents the corresponding time of the nth group of historical data;

[0178] Q3. Zero the mean of the signal to eliminate the influence of the DC component. The expression is as follows:

[0179]

[0180] In the formula, represents the mean value of the input signal X in the mth group of data, Represents the mean value of the input signal Y in the nth group of data;

[0181] Q4. Calculate the correlation coefficient between the input arrays and determine the input signal X m (p) and input signal Y n The correlation between (q) and the data with greater correlation is selected as the input of the MTFE-PatchTST prediction module. The calculation formula is as follows:

[0182] R X'Y' (θ) = ∑[X' m (p)·Y' n (q-θ)]

[0183] In the formula, R X'Y' (θ) represents the cross-correlation function value at the position with lag order θ, R X'Y' The larger the value of (θ), the stronger the cross-correlation.

[0184] Furthermore, the green plant growth correction module is implemented as follows:

[0185] First, various modules are used to collect data of green plants at different growth stages, and K-means is used to aggregate and organize the data. Then, the optimized MTFE-PatchTST is used to make a preliminary prediction of the aggregated data. The actual green plant growth data is subtracted from the preliminary prediction results to obtain the error time series. The mathematical expression of this process is as follows:

[0186] Error=Ovalue-Ivalue

[0187] In the formula, Error represents the error value of the green plant growth data, Ovalue represents the measured value, and Ivalue represents the initial predicted value;

[0188] The error time series is reorganized into input, and then the optimized MTFE-PatchTST model is used to predict the Error using the actual green plant growth data to obtain the error prediction result IError of the green plant growth;

[0189] Finally, the initial prediction result and the error prediction result are added together to get the final prediction result. The formula is as follows:

[0190] Fvalue=Ivalue+IError

[0191] Among them, Fvalue represents the final predicted value of green plant growth.

Claims

1. A green plant growth prediction system based on multi-parameter environmental monitoring, characterized in that: It includes interconnected solar radiation data processing module, pH value data processing module, fertilizer content data processing module, water oxygen content quantification module, cross-correlation function module, green plant growth prediction module and green plant growth correction module; The solar radiation data processing module includes a solar radiation detection module, a solar radiation static data fusion module and a sliding filter solar radiation data processing device; The pH value data processing module includes a pH value detection module, a pH value static data fusion module and a sliding filter pH value data processing device; The fertilizer content data processing module includes a fertilizer content detection module, a fertilizer content static data fusion module and a sliding filter fertilizer content data processing device; The water oxygen content quantification module includes five water oxygen content quantification modules for seedling stage, germination stage, seedling stage, flowering and fruiting stage and vegetative growth stage; The cross-correlation function module is used to measure the similarity between the growth of green plants and solar radiation, fertilizer content in water, oxygen content in water, and pH value of two signals at different time lags; The green plant growth prediction module is an MTFE-PatchTST prediction model; The green plant growth correction module generates a prediction error after the optimized MTFE-PatchTST preliminary prediction, calculates the error sequence, and then uses the optimized MTFE-PatchTST model to correct the error data again.

2. The green plant growth prediction system based on multi-parameter environmental monitoring according to claim 1 is characterized in that: For the solar radiation data processing module, the solar radiation value data of the green plant growth environment is collected through multiple sensors in the solar radiation data processing module, and then the collected pH value is fused and processed through the multi-point solar radiation value static data fusion module, and the solar radiation value after data fusion is output, and then the data abnormal value is processed through the sliding filter solar radiation value data processing device, and the final solar radiation value is output; For the pH value data processing module, the pH value data of the green plant growth environment is collected through multiple sensors in the pH value data processing module, and then the collected pH values ​​are fused and processed through the multi-point pH value static data fusion module, and the pH value after data fusion is output, and then the data abnormal values ​​are processed through the sliding filter pH value data processing device to output the final pH value; For the fertilizer content data processing module, the fertilizer content data of the green plant growth environment is collected through multiple sensors in the fertilizer content data processing module, and then the collected fertilizer content is fused and processed through the multi-point fertilizer content static data fusion module, and the fertilizer content after data fusion is output, and then the data abnormal values ​​are processed through the sliding filter fertilizer content data processing device to output the final fertilizer content; For the water oxygen content quantification module, the fused oxygen content data is quantified by K-means clustering according to 5 different growth stages to obtain the water oxygen content data in different growth stages; The green plant growth prediction module collects and analyzes the historical growth data of green plants and the corresponding historical temperature data, then uses the improved Aurora optimization algorithm to optimize the MTFE-PatchTST scaling factor parameters, and finally uses the optimized MTFE-PatchTST prediction model to perform complex calculations and output prediction results.

3. The detection method of the green plant growth prediction system based on multi-parameter environmental monitoring according to claim 1 is characterized in that: The implementation process of the water oxygen content quantification module based on K-means clustering quantification is as follows: Step (1): randomly select g objects from a oxygen content fusion data as the initial clustering centers; Step (2): Calculate the distance from each data point to the cluster center and assign each data point to the cluster closest to it. The distance is expressed as follows: In the formula, y represents the oxygen content fusion data object, E i represents the i-th cluster center of the oxygen content fusion data cluster, p represents the number of growth periods; y j , E ij Represents y and E i The j-th attribute value of ; Step (3): After all growth periods are assigned, the mean of all growth periods in each cluster is used as the new cluster center of the category to recalculate the distance so that the total square error sum of each sample point to the nearest cluster center becomes smaller. The calculation formula of the total square error sum is as follows; In the formula, the size of the total error square sum indicates the quality of the clustering result, and h represents the number of clusters; Step (4): Compare with the H cluster centers calculated last time. If the cluster centers have changed, go to step (2); otherwise, go to step (5). Step (5): When the cluster center no longer changes, stop and output the cluster quantization results.

4. The detection method of the green plant growth prediction system based on multi-parameter environmental monitoring according to claim 1 is characterized in that: The implementation process of the solar radiation data processing module, the pH value data processing module and the fertilizer content data processing module is as follows: Step (1): Use sensors to collect pH value, solar radiation, and fertilizer data samples in water; Step (2): using static fusion technology to process the sensor's measurement data to obtain processed optimized data; Step (3): Use the sliding filter to process the optimized data to obtain the final data.

5. The detection method of the green plant growth prediction system based on multi-parameter environmental monitoring according to claim 4 is characterized in that: Specifically, Define the final data of each pH sensor, recorded as: y[a] θ1 ; where θ1 is the number of the pH sensor, x[a] θ1 is the input of the pH sliding filter, and a is the number of pH sensors; Define the final data of each solar radiation sensor as: y[b] θ2 ; where θ2 is the number of the pH sensor, x[b] θ2 is the input of the solar radiation sliding filter, b is the number of solar radiation sensors; Define the final data of each fertilizer data sensor, recorded as: y[c] θ3 ; where θ3 is the number of the fertilizer data sensor, x[c] θ3 is the input of the fertilizer data sliding filter, c is the number of fertilizer sensors; The formula of its sliding filter is as follows: Where A is the size of the pH sliding window, B is the size of the solar radiation sliding window, C is the size of the fertilizer sliding window, and k is a positive integer.

6. The detection method of the green plant growth prediction system based on multi-parameter environmental monitoring according to claim 1 is characterized in that: The algorithm improvement implementation steps of the green plant growth prediction module are as follows: Step (1): In the initialization phase, the iterative process of the Aurora optimization algorithm begins with an initial population generated by pseudo-random numbers; this population is represented as a matrix whose dimension is the number of candidate solutions in the population multiplied by the scalable dimension of the solution space, and its formula is as follows: In the Aurora Optimization Algorithm, the search agent simulates the process of high-energy charged particles moving along magnetic field lines toward the Earth's poles in the solution space. The boundaries of the agent are defined by upper and lower bounds, while R represents a random number sequence between 0 and 1; Step (2): During the rotation phase, charged particles approaching the Earth interact with the Earth's magnetic field, causing them to rotate along the magnetic field lines; this phenomenon is described by the formula for the Lorentz force: In the formula, CJ represents the integral constant, Ec represents the charge of the charged particle, Zl represents the mass of the particle, and Qd represents the strength of the geomagnetic field. These four parameters are all fixed values. Therefore, it is assumed that the values ​​of CJ, Ec, and Qd are all equal to 1, and the value of Zl is set to 100. The damping factor Df is randomly selected in the range of 1 to 1.5, and the time is simulated within the strategy using the algorithm's fitness evaluation process. Step (3): In the rotational motion stage, in order to explore the local solution space, find the local optimum or optimize the local structure of the current solution; introduce the aurora ellipse walking strategy to improve local development and fine-tune to obtain the best solution or local optimum; This strategy focuses on global search capability, and particles take larger steps to explore in the solution space; this reflects the global search phase of the algorithm, in which the solution space is extensively searched with larger step sizes; The improved PLO algorithm has the following formula: JZ new (i,j)=X(i,j)+r2×(W1×v(t)+W2×Ao) Ao=Levy(d)×(JZ avg (j)-JZ(i,j))+LB+r1×(UB-LB) / 2 In the formula, JZ new (i,j) represents the position of the high-energy particle after the update, r2 represents the interference of uncontrollable environment and other factors on the particle, and takes the value of [0,1]; Step (4): During the particle collision phase, the chaotic collision between particles enables the PLO to leave the local optimum. When these particles enter the atmosphere and merge with the aurora in the ellipse, the formula is as follows: JZ new (i,j)=JZ(i,j)+sin(r3×π)×(JZ(i,j)-JZ(a,j)),r4<Kandr5<0.05 Where JZ(a,j) refers to a random particle in the particle swarm; as the algorithm progresses, the frequency of collisions between particles gradually increases, and this process is regulated by the collision probability K; the value range of r3 and r4 is [0,1].

7. The detection method of the green plant growth prediction system based on multi-parameter environmental monitoring according to claim 6 is characterized in that: The process of optimizing the scaling factor D of the MTFE-PatchTST model using the improved Aurora optimization algorithm is as follows: Initialize the relevant parameters of the improved Aurora optimization algorithm, including the initial particle position, the upper and lower limits of the search space, the maximum number of iterations, and the current number of iterations; The training set of green plant growth data is input into the MTFE-PatchTST model to optimize the scaling factor D of the model; The new particle position is obtained in the particle collision stage of the improved Aurora optimization algorithm, and the fitness value of each particle is calculated. When the preset termination condition of the model is met, the optimal particle position is determined, which is the optimal value of the scaling factor D of the MTFE-PatchTST model. The optimized MTFE-PatchTST model predicts the data after K-means aggregation and quantization to obtain preliminary green plant growth prediction results.

8. The detection method of the green plant growth prediction system based on multi-parameter environmental monitoring according to claim 1 is characterized in that: The implementation steps of the green plant growth prediction module are as follows: A multivariate time series feature extraction module MTFE consisting of an encoder and a decoder is introduced into the PatchTST model as an extension and improvement of the basic model to form the MTFE-PatchTST model; This extended improvement enhances the model's ability to process data, enabling it to extract features from complex multivariate time series; The specific process is as follows: Step (1): establish a multivariate time series sample set of green plant growth with a lookback window L, where each sample in the sample set contains multiple variables, and the above variables together describe the growth status of the green plants; Each sample can be represented as a multivariate time series, where the length of each variable sequence is L, and the green plant growth sample is set as: t={t1,t2,…,t L }; Step (2): The variable sequence of length L starting from index 1 is expressed as In the formula, k is a positive integer, l specifically refers to the position of the variable in L, and M refers to the total number of variables in the variable sequence; input t = {t1, t2, …, t L } is separated into univariate sequences t (l) ∈R 1×L , R refers to a real number sequence with 1 row and L columns; Among them, each sequence enters the Transformer network according to the channel-independence setting; Step (3): Input the patch-separated historical growth data of green plants into the encoder of the MTFE module and evaluate it through the self-attention mechanism. The evaluation formula is as follows: In the formula, Q is Query, K is Key, V is Value, and D is the scaling factor; Step (4): The output of the encoder is then fed into the MTFE Decoder, which uses a self-attention mechanism to further process the features. Finally, a fully connected layer with a linear head is used to obtain the predicted result, that is, the amount of green plant growth at a certain point in the future:

9. The detection method of the green plant growth prediction system based on multi-parameter environmental monitoring according to claim 1 is characterized in that: The implementation process of the cross-correlation function module is as follows: Step (1): Input discrete signal acquisition data value X m (p), In the formula, m = 1, 2, 3, 4 represent the input oxygen content, pH value, solar radiation and fertilizer data respectively, and p represents the corresponding time of the mth group of data collection; Step (2): Input historical data value Y n (q), In the formula, n=1, 2 represent the input green plant growth and water temperature historical data respectively. q represents the corresponding time of the nth group of historical data; Step (3): Zero-mean the signal to eliminate the influence of the DC component. The expression is as follows: In the formula, represents the mean value of the input signal X in the mth group of data, Represents the mean value of the input signal Y in the nth group of data; Step (4): Calculate the correlation coefficient between the input arrays and determine the input signal X m (p) and input signal Y n (q), select the data with greater correlation as the input of the MTFE-PatchTST prediction module, and its formula is as follows: R X'Y' (θ)=∑[X' m (p)·Y n '(q-θ)] In the formula, R X'Y' (θ) represents the cross-correlation function value at the position with lag order θ, R X'Y' The larger the value of (θ), the stronger the cross-correlation.

10. The detection method of the green plant growth prediction system based on multi-parameter environmental monitoring according to claim 1 is characterized in that: The implementation process of the green plant growth correction module is as follows: First, use each module to collect data of green plants at different growth stages, and use K-means to aggregate and organize the data; Then, the optimized MTFE-PatchTST is used to make a preliminary prediction of the aggregated data; the actual green plant growth data is subtracted from the preliminary prediction results to obtain the error time series; the expression of this process is as follows: Error=Ovalue-Ivalue In the formula, Error represents the error value of the green plant growth data, Ovalue represents the measured value, and Ivalue represents the initial predicted value; The error time series is reorganized into input, and then the optimized MTFE-PatchTST model is used to predict the Error using the actual green plant growth data to obtain the error prediction result IError of the green plant growth; Finally, the initial prediction result and the error prediction result are added together to get the final prediction result, which is as follows: Fvalue=Ivalue+IError Where Fvalue represents the predicted value of the final green plant growth.

Images