A vegetable nutrient solution self-maintenance method and system

By building a self-maintenance network of vegetable nutrient solution and using multi-scale convolutional models for prediction and regulation, the problem of inaccurate regulation of vegetable nutrient solution in the existing technology is solved, and the automation and intelligent self-maintenance of vegetable nutrient solution is realized, and the regulation is improved. Accuracy.

CN114530214BActive Publication Date: 2024-11-01ZHENGZHOU MAIJIA AGRI TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202210140980.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-02-16
Publication Date
2024-11-01
Estimated Expiration
2042-02-16

AI Technical Summary

Technical Problem

The existing technology is difficult to achieve precise regulation and maintenance of vegetable nutrient solution. Due to the low level of artificial regulation, lack of intelligence and automation, and the actual composition of nutrient solution elements has large errors and the standard value, it is difficult to control the influence of external factors.

Method used

By obtaining nutrient solution and environmental information, a self-maintenance network is built to predict future nutrient content, and automated regulation is used for multi-scale convolutional networks and training models to reduce real-time monitoring needs.

Benefits of technology

It realizes the automation and intelligent self-maintenance of vegetable nutrient solution, improves the accuracy of nutrient solution element regulation, and reduces manpower and material consumption.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114530214B_ABST
    Figure CN114530214B_ABST
Patent Text Reader

Abstract

The present invention relates to the technical field of nutrient solution self-maintenance, and specifically to a method and system for self-maintenance of vegetable nutrient solution. The method comprises: obtaining nutrient solution information and environmental information of the nutrient solution at each moment in a historical time period; constructing a nutrient solution self-maintenance network according to the nutrient solution information and environmental information of the culture solution at each moment in the historical time period; training the nutrient solution self-maintenance network according to the nutrient solution information at each moment in the historical time period to obtain a trained nutrient solution self-maintenance network; inputting the nutrient solution information and environmental information of the nutrient solution at the current moment into the trained nutrient solution self-maintenance network, predicting the content of each nutrient element at a future moment, and regulating the nutrient solution using the content of each nutrient element at a future moment. The present invention realizes automated and intelligent self-maintenance of the nutrient solution, and improves the accuracy of regulating each nutrient element in the nutrient solution.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of nutrient solution self-maintenance, and in particular to a vegetable nutrient solution self-maintenance method and system. Background Art

[0002] In recent years, the technology of soilless cultivation of vegetables has been developing continuously, and soilless cultivated vegetables have become the main crops in agriculture. The yield and quality of soilless cultivated vegetables are mainly affected by the ratio of nutrient solution. At present, the maintenance and regulation methods of vegetable nutrient solution mainly rely on human intervention to regulate the ratio of nutrient solution, and the content of nutrient elements in the nutrient solution needs to be monitored in real time; the level of intelligence and automation of human regulation and maintenance of nutrient solution is low, and there is a lack of quantitative indicators; if only the standard value of the standard nutrient solution with fixed concentration is used to regulate the nutrient solution, there will be a large error between the actual composition of the nutrient solution elements and the standard theoretical value; the consumption of the concentration of each element in the nutrient solution will also be affected by external factors such as light, temperature and humidity; therefore, it is difficult to accurately regulate and maintain the nutrient solution. Summary of the invention

[0003] In order to solve the problem that it is difficult to accurately control and maintain the nutrient solution in the prior art, the purpose of the present invention is to provide a vegetable nutrient solution self-maintenance method and system, and the technical solution adopted is as follows:

[0004] In a first aspect, an embodiment of the present invention provides a vegetable nutrient solution self-maintenance method comprising the following steps:

[0005] Obtain the nutrient solution information and environmental information of the nutrient solution at each moment in the historical time period; the nutrient solution information includes pH value, K + Content, Ca 2+ content, content and EC value, K + , Ca 2+ and Recorded as nutrient elements, the environmental information includes light intensity, temperature and humidity;

[0006] According to the nutrient solution information and environmental information of the culture solution at each moment in the historical time period, a nutrient solution self-maintenance network is constructed, and the nutrient solution self-maintenance network is used to predict the content of nutrient elements in the nutrient solution at future moments, which is a multi-scale convolution;

[0007] The nutrient solution self-maintenance network is trained according to the nutrient solution information at each moment in the historical time period to obtain a trained nutrient solution self-maintenance network;

[0008] The nutrient solution information and environmental information of the nutrient solution at the current moment are input into the trained nutrient solution self-maintenance network to predict the content of each nutrient element at the future moment, and the nutrient solution is regulated using the content of each nutrient element at the future moment.

[0009] In a second aspect, another embodiment of the present invention provides a vegetable nutrient solution self-maintenance system, including a memory and a processor, wherein the processor executes a computer program stored in the memory to implement the above-mentioned vegetable nutrient solution self-maintenance method.

[0010] Preferably, the process of constructing a nutrient solution self-maintenance network includes:

[0011] According to the content of each nutrient element in the culture solution at each moment in the historical time period, curve fitting is performed on the content of each nutrient element to obtain a change curve corresponding to the content of each nutrient element, wherein the change curve is used to reflect the change relationship of the corresponding nutrient element content over time;

[0012] Integrate and segment the nutrient solution information and environmental information at the same time to obtain the corresponding data vector;

[0013] According to the variation curves corresponding to the contents of each nutrient element, the length of the time window of the nutrient solution self-maintenance network and the period length of the variation curves corresponding to the contents of each nutrient element are obtained;

[0014] Calculate the expansion coefficient corresponding to each convolutional layer according to the length of the time window;

[0015] The span of the convolution layer is obtained according to the length of the time window and the period length of the change curve corresponding to the content of each nutrient element;

[0016] According to the segmentation of the data vector, a multi-scale convolution kernel is designed, wherein the multi-scale convolution kernel includes a first convolution kernel, a second convolution kernel, and a third convolution kernel; the multi-scale convolution kernel is used to simultaneously convolve a data vector sequence, wherein the data vector sequence is composed of data vectors corresponding to different moments;

[0017] A nutrient solution self-maintenance network is constructed according to the length of the time window, multi-scale convolution kernels, expansion coefficients, and convolution layer span.

[0018] Preferably, the length of the time window of the nutrient solution self-maintenance network is obtained according to the change curve corresponding to the content of each nutrient element, including:

[0019] According to the change curve corresponding to the content of each nutrient element, the slope of the change curve corresponding to the content of each nutrient element when the content of each nutrient element changes with time is obtained;

[0020] The length of the time window is calculated based on the slope of the change curve corresponding to the content of each nutrient element when the content of each nutrient element changes with time.

[0021] Preferably, the convolution layer span is obtained according to the length of the time window and the period length of the change curve corresponding to the content of each nutrient element, including:

[0022] The maximum value of the cycle length of the change curve corresponding to the content of each nutrient element is recorded as the target cycle length;

[0023] The convolutional layer span is calculated based on the target cycle length and the time window length.

[0024] Preferably, the formula for calculating the span of the convolutional layer is:

[0025]

[0026] Among them, K is the span of the convolution layer, K0 is the initial span value of the convolution layer, α is the span adjustment coefficient, To round down;

[0027] The span adjustment coefficient is the ratio of the target cycle length to the length of the time window.

[0028] Preferably, the loss function of the training nutrient solution information on the network is:

[0029]

[0030] Where Loss is the value of the loss function of the nutrient solution self-maintenance network, w is the difference of the environmental feature vector, n is the length of the data vector sequence, is the true value of the content of each nutrient element at the i-th moment, y i It is the predicted value of the content of each nutrient element at the i-th moment output by the nutrient solution self-maintenance network.

[0031] Preferably, the formula for calculating the difference of the environmental feature vector is:

[0032]

[0033] Where m is the total number of target cycles, h j is the environmental feature vector corresponding to the jth target period, h j-1 The environmental feature vector corresponding to the j-1th target period, ‖‖2 is the L2 distance.

[0034] Preferably, the nutrient solution information and environmental information of the nutrient solution at the current moment are input into the trained nutrient solution self-maintenance network, the content of each nutrient element at the future moment is predicted, and the nutrient solution is regulated using the content of each nutrient element at the future moment, including:

[0035] Obtain the nutrient solution information and environmental information of the nutrient solution at the current moment;

[0036] Input the current nutrient solution information and environmental information of the nutrient solution into the trained nutrient solution self-maintenance network to obtain the content of each nutrient element at the future time;

[0037] Calculate the change in the content of each nutrient element based on the content of each nutrient element at a future time and the content of each nutrient element at a current time;

[0038] If the change in the content of each nutrient element is greater than the corresponding change threshold, the corresponding nutrient element content is regulated.

[0039] The embodiments of the present invention have the following beneficial effects:

[0040] The present invention utilizes nutrient solution information and environmental information of the nutrient solution at each moment in a historical time period to construct a nutrient solution self-maintenance network, wherein the nutrient solution self-maintenance network is used to predict the content of nutrient elements in the nutrient solution at future moments, and the nutrient solution self-maintenance network is trained according to the nutrient solution information at each moment in the historical time period to obtain a trained nutrient solution self-maintenance network; the present invention utilizes the trained nutrient solution self-maintenance network to predict the content of each nutrient element at future moments, and then regulates and maintains the nutrient solution, thereby realizing automated and intelligent self-maintenance of the nutrient solution, eliminating the need for real-time monitoring of the content of each element in the nutrient solution, and greatly improving the accuracy of regulating each nutrient element in the nutrient solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0042] Figure 1 The present invention is a flowchart of a vegetable nutrient solution self-maintenance method provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0043] In order to further explain the technical means and functional effects adopted by the present invention to achieve the predetermined purpose of the invention, a vegetable nutrient solution self-maintenance method and system proposed in accordance with the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0044] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0045] The specific scheme of the vegetable nutrient solution self-maintenance method and system provided by the present invention is described in detail below with reference to the accompanying drawings.

[0046] A vegetable nutrient solution self-maintenance method embodiment:

[0047] like Figure 1 As shown, a vegetable nutrient solution self-maintenance method of this embodiment includes the following steps:

[0048] Step S1, obtaining nutrient solution information and environmental information of the nutrient solution at each time in the historical time period; the nutrient solution information includes pH value, K + Content, Ca 2+ content, content and EC value, K + , Ca 2+ and Recorded as nutrient elements, the environmental information includes light intensity, temperature and humidity.

[0049] The scenario targeted by this embodiment is a soilless vegetable culture scenario, and there is a corresponding data acquisition device in the culture environment. The data acquisition device can obtain nutrient solution information in the soilless culture medium through a sensor, that is, nutrient solution information in the nutrient solution, and obtain environmental information of vegetable culture; the nutrient solution information includes pH value, K + Content, Ca 2+ content, content and EC value (used to measure the concentration of soluble salts in the solution), the environmental information includes: light intensity, temperature and humidity. In this embodiment, K + , Ca 2+ and The present embodiment is described by taking the scenario of soilless cultivation of tomatoes as an example.

[0050] This embodiment uses a data acquisition device to collect nutrient solution information and environmental information of the soilless culture medium at each collection time in a historical time period. In this embodiment, the historical time period is 3 months, and the collection time interval is 1 hour, which can be set according to actual needs. This embodiment uses the nutrient solution information at each time in the historical time period to collect pH value, K + Content, Ca 2+ content, The content and EC value are curve fitted to obtain the change curve corresponding to each information in the nutrient solution information. In this embodiment, the missing data value can be processed by fitting the curve to obtain the change relationship between each data in the nutrient solution information and time.

[0051] The nutrient solution data and environmental information collected in this embodiment are used to train the nutrient solution self-maintenance network constructed subsequently.

[0052] Step S2, constructing a nutrient solution self-maintenance network based on the nutrient solution information and environmental information of the culture solution at each moment in the historical time period, wherein the nutrient solution self-maintenance network is used to predict the content of nutrient elements in the nutrient solution at future moments, and is a multi-scale convolution.

[0053] This embodiment constructs a nutrient solution self-maintenance network based on the information types of the nutrient solution information and the environmental information of the culture solution. The input of the nutrient solution self-maintenance network is a data sequence consisting of the nutrient solution information and the environmental information at a certain moment, and the output is the nutrient element content at a future moment. That is, the nutrient solution self-maintenance network is used to predict the nutrient element content of the nutrient solution at a future moment, so as to realize the regulation of the nutrient solution.

[0054] In this embodiment, the nutrient solution self-maintenance network is an Encoder-FC structure. The training data of the nutrient solution self-maintenance network is the nutrient solution information and environmental information at different times in the historical time period. In this embodiment, the nutrient solution information and environmental information at different times in the historical time period are input into the nutrient solution self-maintenance network to train the nutrient solution self-maintenance network. Next, this embodiment describes the design of the internal structure of the nutrient solution self-maintenance network based on the nutrient solution information and environmental information:

[0055] Since the training data at a moment is multidimensional data, including environmental information and nutrient solution information, and there is a certain relationship between the data dimensions, in order to enable the network to quickly extract features, this embodiment preprocesses the data input to the network before convolution, integrates the multidimensional data at the same moment, and divides it to obtain a data vector corresponding to the moment, that is, the data corresponding to one moment corresponds to one data vector, so as to facilitate the subsequent multi-scale convolution of the data vector, specifically: [[A, B, C], [D, E], [F, G, H]], where A represents the K of the nutrient solution + Content, B represents the Ca content of the nutrient solution 2+ Content, C represents the content of nutrient solution content (this embodiment designs the nutrient solution self-maintenance network based on the segmentation of the data vector, that is, the ratio and composition of the nutrient solution), D represents the pH value of the nutrient solution, E represents the EC value of the nutrient solution, F represents the light intensity, G represents the temperature, and H represents the humidity.

[0056] After preprocessing the data, this embodiment uses a multi-scale convolution kernel to simultaneously convolve the data vector sequence to obtain the corresponding data features. The data vector sequence is a sequence composed of all data vectors in the historical time period. Traditional time series prediction networks use data from adjacent moments and previous moments to predict data at future moments. The time window constraint is small, and for a large amount of data information, many convolution layers need to be stacked to obtain the final output. This embodiment designs an adaptive time window length L (i.e., convolution kernel size), expansion coefficient d, multi-scale convolution, and convolution layer span K to reduce the number of stacked convolution layers and reduce network overhead. Specifically:

[0057] First, design an adaptive time window length.

[0058] This embodiment determines the length of time when the slopes are consistent by judging the consistency of the slopes of the change curves corresponding to A, B, and C obtained in step S1, that is, judging whether the positive and negative slopes of the change curves corresponding to the content of each nutrient element are consistent within a period of time. Specifically:

[0059] Obtain the slope k of the change curve corresponding to the content of each nutrient element when the content of each nutrient element changes over time A ,k B ,k C , where k A K + The slope corresponding to the content, k B Ca 2+ The slope corresponding to the content, k C for The slope corresponding to the content; if k A ,k B ,k C >0 or k A ,k B ,k C <0, it is considered that the three nutrient elements have slope consistency during this period of time; this embodiment determines the length of time for each slope consistency based on the slope consistency of the change curves corresponding to the three nutrient elements, that is, the length of time from the beginning of any slope consistency to the end of the slope consistency.

[0060] According to the above method, this embodiment can obtain multiple time lengths in the historical time period to form a corresponding time length sequence [l1, l2, ..., l m ], where l1 is the length of time when the slope is consistent for the first time, l2 is the length of time when the slope is consistent for the second time, and l m is the time length with the same slope for the mth time; this embodiment uses the time length sequence to obtain the mean time length, and uses the mean time length as the time window length, that is, Where L is the time window length, l i is the time length with consistent slope for the i-th time, and m is the number of time lengths in the time length sequence.

[0061] Second, design the expansion factor.

[0062] The change of the expansion coefficient is only related to the size of the convolution window (i.e., the time window). The larger the convolution window, the larger the data interval in the convolution window. The data interval size is the expansion coefficient. In this embodiment, the expansion coefficient d=L n , where d is the expansion coefficient of the nth convolutional layer.

[0063] Third, design the span of the convolutional layer.

[0064] In this embodiment, the periodicity of the change curves corresponding to A, B, and C is determined respectively, and the period length Z of the change curve of the content of each nutrient element is obtained when each nutrient element changes with time. A ,Z B ,Z C , where Z A K + The period length of the change curve corresponding to the content, Z B Ca 2+ The period length of the change curve corresponding to the content, Z C for The period length of the variation curve corresponding to the content. In this embodiment, a period length may include multiple time window lengths.

[0065] This embodiment transfers cross-layer information between convolutional layers according to the change in the period length of the change curve corresponding to the content of the three nutrient elements. This embodiment selects the maximum value max(Z A ,Z B ,Z C ), recorded as the target cycle length, the corresponding cycle is the target cycle, and then the ratio of the target cycle length to the time window length obtained above is used as the span adjustment coefficient, that is, In this embodiment, the span adjustment coefficient is used to control the span of cross-layer information transmission of the convolution layer. In this embodiment, the longer the target cycle length is, the more time windows are included in the target cycle, that is, the more data changes, the smaller the span between the convolution layers should be, in order to better learn the changing rules of the nutrient content in the target cycle; the shorter the target cycle length is, the fewer time windows are included in the target cycle, the larger the span between the convolution layers should be, and the changing rules of the nutrient solution elements between cycles can be better learned. The calculation formula for obtaining the span of the convolution layer in this embodiment is:

[0066]

[0067] Among them, K is the span of the convolution layer, K0 is the initial span value of the convolution layer, α is the span adjustment coefficient, It represents rounding down. In this embodiment, K0=5, which can be adjusted according to actual needs.

[0068] Fourth, design multi-scale convolution kernels and convolution process.

[0069] This embodiment takes into account that the prediction of the content of each nutrient element in the nutrient solution is not only related to the time series, but also related to the environmental information, pH value and EC value.

[0070] This embodiment uses a multi-scale convolution kernel to simultaneously convolve the input data vector sequence to extract corresponding features. In this implementation, two scales of convolution kernels are set in the same convolution layer to convolve the data vector sequence. The sizes of the two scales of convolution kernels are: [1,3] and [1,2], i.e., 1 row and 3 columns and 1 row and 2 columns.

[0071] This embodiment designs three convolution kernels according to the constituent components of the data vector, which are respectively recorded as the first convolution kernel, the second convolution kernel and the third convolution kernel, and the scales of the three convolution kernels are [1,3], [1,2], and [1,3] respectively; this embodiment uses these three different convolution kernels to simultaneously convolve the data vector sequence, and the convolution process is: the three convolution kernels slide within the time window and convolve different data. The first convolution kernel obtains the characteristics of the three nutrient elements within the time window, the second convolution kernel obtains the characteristics of the pH value and the EC value within the time window, and the third convolution kernel obtains the characteristics of the environmental information within the time window. However, the sliding step size of the third convolution kernel is different from that of the first convolution kernel and the second convolution kernel. The third convolution kernel does not update the weight parameters, that is, the characteristics obtained by the third convolution kernel remain unchanged in a target period, and will only change when sliding to the next target period.

[0072] When the length of the sliding time window of the convolution kernel is greater than or equal to the target cycle length, the third convolution kernel begins to convolve the environmental information in the first time window in the previous target cycle to obtain the environmental feature vector corresponding to the target cycle, and then stabilizes the features obtained by the third convolution kernel in the sliding process of the time window under the next target cycle length, and continues this process until the first layer of convolution ends. Then, the next layer of convolution is performed using the convolution layer span, and iterates continuously. As the number of iterations increases, the receptive field of the convolution kernel continues to expand. The deeper the convolution, the more target cycle lengths it may contain. At this time, the third convolution kernel performs sliding convolution between cycles, and the data within the cycle is regarded as a hole until the data feature convolution is completed. The convolution kernel data convolution process in this embodiment is a prior art and will not be repeated here.

[0073] As another embodiment, if the nutrient solution contains four nutrient elements, such as K + , Ca 2+ , While other data remain unchanged, three scales of convolution kernels are set in the same convolution layer to convolve the data vector sequence. The sizes of the three scales of convolution kernels are: [1,4], [1,3] and [1,2], respectively. The designed convolution kernels are: the scale of the first convolution kernel is [1,4], which is used to convolve the nutrient element data; the scale of the second convolution kernel is [1,2], which is used to convolve the pH value and EC value; the scale of the third convolution kernel is [1,3], which is used to convolve the environmental information; that is, the scale of the convolution kernel and the number of convolution kernels in this embodiment should be designed according to actual needs.

[0074] This embodiment designs a convolution method of a multi-scale convolution kernel according to the characteristics of the nutrient solution data. It can obtain the changing characteristics of environmental parameters between cycles according to different cycle lengths, and can also obtain the changing characteristics of nutrient solution elements within a cycle when the characteristic parameters of the environmental parameters remain unchanged, thereby making the prediction results more accurate.

[0075] Step S3, training the nutrient solution self-maintenance network according to the nutrient solution information at each moment in the historical time period to obtain a trained nutrient solution self-maintenance network.

[0076] In this embodiment, step S2 designs the nutrient solution self-maintenance network according to the change curve of the content of each nutrient element in the nutrient solution within the historical time period. Furthermore, this embodiment uses the nutrient solution information and environmental information of the nutrient solution at different times within the historical time period to train the nutrient solution self-maintenance network. Specifically, the nutrient solution information and environmental information of the nutrient solution at different times within the historical time period are first input into the nutrient solution self-maintenance network; then the network integrates the nutrient solution information and environmental information of the nutrient solution at the same time within the historical time period to obtain data vectors at different times; finally, the first convolution kernel, the second convolution kernel and the third convolution kernel are used to simultaneously convolve the data vector sequence; during the network training process, the network parameters will change dynamically with the changes in the network input data until the network converges to obtain a trained nutrient solution self-maintenance network.

[0077] The formula for the loss function of training the nutrient solution self-maintenance network in this embodiment is:

[0078]

[0079] Where Loss is the value of the loss function of the nutrient solution self-maintenance network, w is the difference of the environmental feature vector, and n is the length of the data vector sequence, that is, the length of the time series within the historical time end, a total of n moments. is the true value of the content of each nutrient element at the i-th moment, y i It is the predicted value of the content of each nutrient element at the i-th moment output by the nutrient solution self-maintenance network.

[0080] The difference of the environmental feature vector is used to reflect the environmental changes during the target period. The greater the difference of the environmental feature vector, the greater the environmental changes, and the predicted results may change to a certain extent. The formula for calculating the difference of the environmental feature vector in this embodiment is:

[0081]

[0082] Where m is the total number of target cycles, h j is the environmental feature vector corresponding to the jth target period, h j-1 The environmental feature vector corresponding to the j-1th target period, ‖‖2 is the L2 distance.

[0083] Step S4, inputting the current nutrient solution information and environmental information of the nutrient solution into the trained nutrient solution self-maintenance network, predicting the content of each nutrient element at a future time, and using the content of each nutrient element at a future time to regulate the nutrient solution.

[0084] Next, this embodiment predicts the content of each nutrient element in the nutrient solution through the trained nutrient solution self-maintenance network, and adjusts the nutrient solution according to the predicted value. Specifically:

[0085] Obtain the nutrient solution information and environmental information of the nutrient solution at the current moment, and then input the nutrient solution information and environmental information of the nutrient solution at the current moment into the trained nutrient solution self-maintenance network to obtain the content of each nutrient element at the future moment, that is, K at the future moment + Content, Ca 2+ Content and Then, according to the content of each nutrient element at the future moment and the content of each nutrient element at the current moment, the change in the content of each nutrient element is calculated, that is, the difference between the content of the nutrient element at the current moment and the content at the future moment is calculated.

[0086] In this embodiment, a corresponding change threshold is set for each nutrient element. When it is predicted that the change in the content of each nutrient element is greater than the corresponding change threshold, the corresponding nutrient element content is regulated to achieve self-maintenance of the nutrient solution. There is no need to monitor the content of each element in the nutrient solution in real time, which also saves a lot of manpower and material resources. The threshold in this embodiment needs to be set according to actual needs.

[0087] This embodiment uses the nutrient solution information and environmental information of the nutrient solution at each moment in the historical time period to construct a nutrient solution self-maintenance network, and the nutrient solution self-maintenance network is used to predict the content of nutrient elements in the nutrient solution at future moments, and train the nutrient solution self-maintenance network according to the nutrient solution information at each moment in the historical time period to obtain a trained nutrient solution self-maintenance network; this embodiment uses the trained nutrient solution self-maintenance network to predict the content of each nutrient element at future moments, and then regulates and maintains the nutrient solution, thereby realizing automated and intelligent self-maintenance of the nutrient solution, without the need for real-time monitoring of the content of each element in the nutrient solution, and also takes into account the influence of the external environment on the content of each element in the nutrient solution, thereby greatly improving the accuracy of regulation of each nutrient element in the nutrient solution.

[0088] Embodiment of a vegetable nutrient solution self-maintenance system

[0089] A vegetable nutrient solution self-maintenance system in this embodiment includes a memory and a processor, and the processor executes a computer program stored in the memory to implement the above-mentioned vegetable nutrient solution self-maintenance method.

[0090] Since a method for self-maintenance of a vegetable nutrient solution has been described in an embodiment of a method for self-maintenance of a vegetable nutrient solution, this embodiment will not further describe the method for self-maintenance of a vegetable nutrient solution.

[0091] It should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A vegetable nutrient solution self-maintenance method, characterized in that: The method comprises the following steps: Obtain the nutrient solution information and environmental information of the nutrient solution at each moment in the historical time period; the nutrient solution information includes pH value, K + Content, Ca 2+ content, content and EC value, K + , Ca 2+ and Recorded as nutrient elements, the environmental information includes light intensity, temperature and humidity; According to the nutrient solution information and environmental information of the culture solution at each moment in the historical time period, a nutrient solution self-maintenance network is constructed, and the nutrient solution self-maintenance network is used to predict the content of nutrient elements in the nutrient solution at future moments, which is a multi-scale convolution; The nutrient solution self-maintenance network is trained according to the nutrient solution information at each moment in the historical time period to obtain a trained nutrient solution self-maintenance network; Input the current nutrient solution information and environmental information into the trained nutrient solution self-maintenance network, predict the content of each nutrient element at the future time, and use the content of each nutrient element at the future time to regulate the nutrient solution, including: Obtain the nutrient solution information and environmental information of the nutrient solution at the current moment; Input the current nutrient solution information and environmental information of the nutrient solution into the trained nutrient solution self-maintenance network to obtain the content of each nutrient element at the future time; Calculate the change in the content of each nutrient element based on the content of each nutrient element at a future time and the content of each nutrient element at a current time; If the change in the content of each nutrient element is greater than the corresponding change threshold, the corresponding nutrient element content is regulated; The process of building a nutrient solution self-maintenance network includes: According to the content of each nutrient element in the culture solution at each moment in the historical time period, curve fitting is performed on the content of each nutrient element to obtain a change curve corresponding to the content of each nutrient element, wherein the change curve is used to reflect the change relationship of the corresponding nutrient element content over time; Integrate and segment the nutrient solution information and environmental information at the same time to obtain the corresponding data vector; According to the variation curves corresponding to the contents of each nutrient element, the length of the time window of the nutrient solution self-maintenance network and the period length of the variation curves corresponding to the contents of each nutrient element are obtained; Calculate the expansion coefficient corresponding to each convolutional layer according to the length of the time window; The span of the convolution layer is obtained according to the length of the time window and the period length of the change curve corresponding to the content of each nutrient element; According to the segmentation of the data vector, a multi-scale convolution kernel is designed, wherein the multi-scale convolution kernel includes a first convolution kernel, a second convolution kernel, and a third convolution kernel; the multi-scale convolution kernel is used to simultaneously convolve a data vector sequence, wherein the data vector sequence is composed of data vectors corresponding to different moments; A nutrient solution self-maintenance network is constructed according to the length of the time window, multi-scale convolution kernels, expansion coefficients, and convolution layer span.

2. A vegetable nutrient solution self-maintenance method according to claim 1, characterized in that: According to the change curve corresponding to the content of each nutrient element, the length of the time window of the nutrient solution self-maintenance network is obtained, including: According to the change curve corresponding to the content of each nutrient element, the slope of the change curve corresponding to the content of each nutrient element when the content of each nutrient element changes with time is obtained; The length of the time window is calculated based on the slope of the change curve corresponding to the content of each nutrient element when the content of each nutrient element changes with time.

3. A vegetable nutrient solution self-maintenance method according to claim 1, characterized in that: According to the length of the time window and the period length of the change curve corresponding to the content of each nutrient element, the span of the convolution layer is obtained, including: The maximum value of the cycle length of the change curve corresponding to the content of each nutrient element is recorded as the target cycle length; The convolutional layer span is calculated based on the target cycle length and the time window length.

4. A vegetable nutrient solution self-maintenance method according to claim 3, characterized in that: The formula for calculating the span of the convolutional layer is: Among them, K is the span of the convolution layer, K0 is the initial span value of the convolution layer, α is the span adjustment coefficient, To round down; The span adjustment coefficient is the ratio of the target cycle length to the length of the time window.

5. A vegetable nutrient solution self-maintenance method according to claim 1, characterized in that: The loss function of the training nutrient solution information on the network is: Where Loss is the value of the loss function of the nutrient solution self-maintenance network, w is the difference of the environmental feature vector, n is the length of the data vector sequence, is the true value of the content of each nutrient element at the i-th moment, y i It is the predicted value of the content of each nutrient element at the i-th moment output by the nutrient solution self-maintenance network.

6. A vegetable nutrient solution self-maintenance method according to claim 5, characterized in that: The formula for calculating the difference of the environmental feature vector is: Where m is the total number of target cycles, h j is the environmental feature vector corresponding to the jth target period, h j-1 The environmental feature vector corresponding to the j-1th target period, ‖‖2 is the L2 distance.

7. A vegetable nutrient solution self-maintenance system, comprising a memory and a processor, characterized in that: The processor executes the computer program stored in the memory to implement a vegetable nutrient solution self-maintenance method as described in any one of claims 1-6.

Citation Information

Patent Citations

  • Water and fertilizer mixing device and water and fertilizer mixing method

    CN110419437A

  • Intelligent water and fertilizer equipment based on soil crop interaction model and optimization regulation and control method thereof

    CN113597864A