CN117272851B - Modeling prediction method for plant light receiving quantity under saline-alkali soil photovoltaic panel - Google Patents

Modeling prediction method for plant light receiving quantity under saline-alkali soil photovoltaic panel Download PDF

Info

Publication number
CN117272851B
CN117272851B CN202311574449.4A CN202311574449A CN117272851B CN 117272851 B CN117272851 B CN 117272851B CN 202311574449 A CN202311574449 A CN 202311574449A CN 117272851 B CN117272851 B CN 117272851B
Authority
CN
China
Prior art keywords
data
light quantity
model
prediction
sequence
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202311574449.4A
Other languages
Chinese (zh)
Other versions
CN117272851A (en
Inventor
任密蜂
牛晓云
封晓辉
孙芳
李玉洁
张博
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Taiyuan University of Technology
Original Assignee
Taiyuan University of Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Taiyuan University of Technology filed Critical Taiyuan University of Technology
Priority to CN202311574449.4A priority Critical patent/CN117272851B/en
Publication of CN117272851A publication Critical patent/CN117272851A/en
Application granted granted Critical
Publication of CN117272851B publication Critical patent/CN117272851B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/10—Geometric CAD
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G06N3/0455—Auto-encoder networks; Encoder-decoder networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/02—Knowledge representation; Symbolic representation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Geometry (AREA)
  • Business, Economics & Management (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Computer Hardware Design (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Strategic Management (AREA)
  • Biophysics (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Development Economics (AREA)
  • Mathematical Optimization (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Pure & Applied Mathematics (AREA)
  • Computational Mathematics (AREA)
  • Game Theory and Decision Science (AREA)
  • Mathematical Analysis (AREA)
  • Medical Informatics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention provides a modeling prediction method for plant received light quantity under a saline-alkali soil photovoltaic panel, belonging to the technical field of modeling prediction for plant received light quantity; the technical problems to be solved are as follows: the improvement of a modeling and predicting method for plant received light quantity under a saline-alkali soil photovoltaic panel is provided; the technical scheme adopted for solving the technical problems is as follows: collecting photosynthetic effective radiation, air temperature and humidity data in a target area by adopting instrument equipment, and continuously collecting annual photosynthetic effective radiation instantaneous data at a plurality of different positions between two adjacent rows of photovoltaic panels paved in a saline-alkali soil at preset time intervals; simulating and analyzing the illumination condition of the missing value in the simulation data set, supplementing the missing value by using a homogeneous mean interpolation method, and converting the instantaneous quantity of the photosynthetic active radiation into an effective photosynthetic accumulation quantity by adopting an integral mode; constructing a loss function guidance training model with priori knowledge constraint based on expert experience; the method is applied to prediction of the received light quantity of plants.

Description

Modeling prediction method for plant light receiving quantity under saline-alkali soil photovoltaic panel
Technical Field
The invention provides a modeling prediction method for plant received light quantity under a saline-alkali soil photovoltaic panel, and belongs to the technical field of modeling prediction for plant received light quantity.
Background
Along with the continuous development of social economy, the technology support is provided for the agricultural industry practice by depending on the technologies such as the Internet, big data, neural network model and the like so as to further improve the agricultural production efficiency; the method can help achieve the double-carbon target aiming at the development mode of the saline-alkali soil and the photovoltaic, improve the utilization rate of the saline-alkali soil and ensure the grain safety.
At present, aiming at the management of agricultural production of saline-alkali soil, the problems of irregular distribution of plant receiving light quantity under the plate caused by paving a photovoltaic plate on the saline-alkali soil and undefined illumination-water-salt dynamic mechanism in a photovoltaic ecological system of the saline-alkali soil exist, so that soil water evaporation and salt migration are influenced, the influence of the construction of a scientific evaluation photovoltaic array on the ecology of the saline-alkali soil can be selected for subsequent crops, effective theoretical guidance is provided for expanding available resources, and the key problem is that the method aims at the research on the change of growth factors under the photovoltaic plate, and particularly relates to the prediction of the receiving light quantity of the plant growing under the photovoltaic plate; at present, the research on various environmental factors under a photovoltaic panel is mostly based on an in-situ ecological experiment, and the research is focused on the influence of the change of the environmental factors on the growth and development of plants, no deep systematic analysis is performed, the analysis lacks theoretical basis and is not objective, the modeling and prediction of the received light quantity of the growing plants under the photovoltaic panel are performed by using a deep learning method, the physical characteristics of data signals are not needed to be understood deeply, and good precision can be obtained, but the prediction and evaluation results are not ideal due to the uncontrollable training process and the unexplained model, and the improvement is needed.
Disclosure of Invention
The invention aims to overcome the defects in the prior art, and solves the technical problems that: aiming at the problems that deep systematic analysis is not carried out on the research of various environmental factors under the photovoltaic panel, the used deep learning method cannot be controlled in the training process and the prediction evaluation result is not ideal due to the unexplainability of the model, the modeling prediction method for the plant received light quantity under the photovoltaic panel of the saline-alkali soil is provided.
In order to solve the technical problems, the invention adopts the following technical scheme: a modeling and predicting method for plant light receiving quantity under a saline-alkali soil photovoltaic panel comprises the following modeling and predicting steps:
step one: collecting photosynthetic effective radiation, air temperature and humidity data in a target area by adopting instrument equipment, continuously collecting annual photosynthetic effective radiation instantaneous data at a plurality of different positions between two adjacent rows of photovoltaic panels paved in a saline-alkali soil at preset time intervals, and establishing a data set;
step two: simulating and analyzing the illumination condition of the missing values in the data set, supplementing the missing values by using a homogeneous mean interpolation method, converting the instantaneous quantity of photosynthetically active radiation into an effective photosynthetically accumulation quantity by adopting an integral mode, and establishing a model;
step three: constructing a loss function with priori knowledge constraint based on expert experience, guiding a model training process, providing useful priori information under the condition of limited data or noise of the data, and limiting the predicted output of the model to a specified range;
step four: an encoder-decoder prediction model is constructed on the basis of a long-term memory neural network layer with a loss function constrained by priori knowledge aiming at an effective light quantity prediction data design algorithm of a position in a target area, and a mapping relation from a historical effective light quantity sequence to a future effective light quantity sequence is trained to realize effective light quantity prediction based on historical data;
and applying a design algorithm to other positions to respectively establish an effective light quantity prediction model, and obtaining the future light quantity receiving condition of the whole saline-alkali soil under the same condition based on the prediction model.
The specific method for simulating and analyzing the illumination condition of the missing value in the data set in the second step is as follows:
introducing weather data of a place where a saline-alkali soil photovoltaic test base is located into Ecotect software, establishing a 3D simulation model of a photovoltaic panel array, simulating shading conditions of each measured position, and obtaining simulation data of photosynthetic effective radiation;
multiplying photosynthetic effective radiation data by the average plant occupied area, and integrating the time period from 0 to t to obtain the photosynthetic effective accumulation amount received in the time period from 0 to t of the position;
superposing the effective light quantity in one day to obtain the energy obtained after the position is subjected to continuous change of illumination shading conditions, loading data, converting the data format into an array, carrying out maximum and minimum normalization on the data, scaling the data to a range from 0 to 1, setting a time window step length n+m, a window moving step length 1, slicing the data into samples, and calculating the formula:
number of samples = data amount- (n+m);
wherein n is a historical time step for training, each time step comprises three input features, namely effective light quantity, highest air temperature and average air humidity of the same day, m is a time span of future data to be predicted, and the unit is days;
in order to improve the generalization performance of the model, samples are randomly disturbed, and a sample set is subjected to 7:2:1 is divided into a training set, a verification set and a test set;
selecting effective light quantity for n consecutive daysX=[E 1 ,…,E n ]As an input sequence of the model,h=[E n+1 ,…,E n+m ]as an output sequence of the model.
The specific method for constructing the loss function with priori knowledge constraint in the third step is as follows:
determining a priori constraint rule by determining a value range of an output value of a model and a change rate range of the output value, and defining two constraint rules, wherein the two constraint rules are specifically as follows:
constraint rule 1: y is∈[34641.6,440714.4]；
Constraint rule 2: rate of change= | [ y (t+1) -y (t)]/y(t)|∈[0.05%,567%]；
Wherein y (t) is the predicted result of the effective light quantity model on the t th day after normalization;
the built custom loss function comprises: average absolute error functionMAEA loss caused by constraint rule 1, a loss caused by constraint rule 2, wherein the average absolute error function is selectedMAEThe calculation formula of (2) is as follows:
；
in the method, in the process of the invention,MAErepresents the average of the absolute error between the predicted value and the observed value,y i for the predicted values at different times for each sample,y ture for the true value of its corresponding instant in time,mfor predicting a time step.
The specific method for predicting by the effective light quantity prediction model in the fourth step is as follows:
the built encoder-decoder model comprises 3 long-period and short-period memory network layers and 1 full-connection layer, the data samples are trained through the model, the data samples are input into the encoder in sequence, the loss function with priori knowledge constraint and the neural network layer with long-period memory are fused for forgetting, memorizing and learning, and the hidden state at the current moment is adoptedh t Is hidden from the last momenth t-1 And current inputx t The common influence satisfies the expression:
h t =f(h t-1 ,X t )；
wherein,ffor the selected activation function, in particular the ReLu function;
the input sequence is then compiled into a semantic vector c containing contextual characteristics, the semantic vector c being defined in particular by the state of the final hidden layerh n The calculation formula is as follows:
c=q(h n )；
the semantic vector c is taken as the initial state of the decoding stage of the decoder, and the state of the last hidden layerh n Decoding as a decoderInitial input of phasey 0 ；
Using the semantic vector c as the input of each neuron of the decoder to set the hidden state at the time th t Is hidden from the last momenth t-1 And output at the last timey t-1 And semantic sequence c effects, expressed as:
h t =f(h t-1 ,y t-1 ,c)；
after decoding different features of the semantic vector c, the decoder reconstructs the label corresponding to the sample, and finally, the sequence-to-sequence prediction is realized through the full connection layer.
Compared with the prior art, the invention has the following beneficial effects: the plant received light quantity modeling prediction method provided by the invention adopts a PK-LSTM layer to establish an encoder-decoder time sequence model, an optimizer with optimal learning rate is used for adjusting the connection weight between multiple layers of neurons in a neural network model to minimize a loss function, the loss function is constructed based on expert experience, the model training process is guided by the loss function with priori knowledge, the model interpretation is increased, historical effective light quantity time sequence data is encoded into an upper Wen Yuyi vector and a lower Wen Yuyi vector by an encoder, the semantic vector is reconstructed by a decoder, and finally the prediction from an effective light quantity sequence to a sequence is realized; according to the invention, the middle-long term prediction model obtained through the deep neural network model training can limit the prediction output of the model within a specified range, so that the prediction result has stronger robustness; the model established by the invention can predict the effective light quantity of multiple days in the future based on the light quantity history data, evaluate the influence of the photosynthetic effective radiation received by the plant under the plate by constructing the photovoltaic array in the saline-alkali soil, and provide effective theoretical guidance for the subsequent crop selection and the development of available resources.
Drawings
The invention is further described below with reference to the accompanying drawings:
FIG. 1 is a flow chart of steps of a modeling prediction method of the present invention;
FIG. 2 is a schematic structural diagram of a model for predicting the effective light quantity of the photovoltaic PK-LSTM in the saline-alkali soil;
FIG. 3 is a graph showing the statistical effects of the collected data set and range characteristics of the present invention;
FIG. 4 is a graph of the performance effects of the model validation set of the present invention;
FIG. 5 is a graph showing the performance effects of the model test set of the present invention;
FIG. 6 is a graph showing the effect distribution of the test distributed points of the prediction model of the present invention under the constraint and unconstrained conditions.
Detailed Description
As shown in FIG. 1, the invention provides a modeling prediction method of plant received light quantity, which is mainly used for predicting the light quantity receiving condition of plants under a saline-alkali soil photovoltaic panel based on a deep neural network prediction model, and particularly relates to a plant growth effective light quantity prediction method of a Long Short-term Memory (LSTM) neural network by integrating priori knowledge; in order to realize the prediction method, the invention adopts a saline-alkali soil-photovoltaic plant growth effective light quantity depth neural network prediction model fused with priori knowledge, the model utilizes expert experience to construct a loss function, guides the training process of the model, uses the constructed model to establish the mapping relation from an effective light quantity history sequence to a prediction sequence to realize multi-step prediction, and finally the established distribution condition model of the under-board illumination condition can better simulate the prediction light quantity sequence conforming to a specific rule, limits the prediction output of the model in a reasonable range and has better capability of treating abnormal conditions.
The deep neural network prediction model used in the invention takes an encoder-decoder network as a basic framework of the model, and uses a PK-LSTM neural network layer to replace a traditional cyclic neural network as an encoder and a decoder, so that the overfitting phenomenon can be prevented, the whole encoder-decoder neural network contains 200 neurons in total, wherein an activation function selects a Relu function, and an Adam optimizer with a learning rate of 0.035 is used.
The prediction model constructed by the invention is particularly a prediction model of an Encoder-Decoder (LSTM) integrating priori knowledge and Long-term Memory, and the problem that the distribution of the plant received light quantity under a plate is irregular and the illumination-water-salt dynamic mechanism in a saline-alkali soil photovoltaic ecological system is undefined after the photovoltaic plate is paved in the saline-alkali soil can be solved by predicting the plant received light quantity through the model; the main steps adopted include:
collecting photosynthetic effective radiation, air temperature, humidity and other data of different positions among photovoltaic plates paved on the saline-alkali soil;
preprocessing the acquired data;
constructing a loss function by using expert experience, and guiding a model training process;
an encoder-decoder model is built using LSTM layers.
Furthermore, the photovoltaic plant growth effective light quantity prediction method based on the depth neural network fusion priori knowledge provided by the invention specifically comprises the following steps:
step one: as shown in fig. 3, special instruments and equipment are adopted to collect photosynthetic effective radiation, air temperature, humidity and other data, and annual photosynthetic effective radiation instantaneous data of 5 different positions between two rows of adjacent photovoltaic panels laid in saline-alkali soil are continuously collected at intervals of 15 minutes, wherein the units are as follows: umol x m -2 *s -1 ；
In the embodiment of the invention, a photosynthetically active radiation sensor arranged in the field detects and collects a photosynthetically active radiation data set, a photovoltaic plate array is paved on the saline-alkali soil, the plate spacing is 11 meters, one sensor is arranged every 2 meters from 1 meter, 5 sensors are arranged between two rows of photovoltaic plates, and the photosynthetically active radiation instantaneous quantity of different positions between the plates and under the plate under the shading condition of the photovoltaic plates is respectively measured.
Step two: simulating and analyzing the illumination condition of the missing value in the data set by using Ecotect software, supplementing the missing value by using a homogeneous mean interpolation method, and converting the instantaneous quantity of photosynthetically active radiation into an effective photosynthetically accumulation quantity (effective light quantity) by using an integral idea;
the method adopts Ecotect software to combine with the similar mean interpolation method to process the data missing value of the original data set, comprises data on-site weather condition importing software, simulates the real-time illumination condition of missing data, and takes the mean value of non-missing data at the same illumination condition as the missing data to correct the missing data.
The invention adopts an illumination environment platform provided by Ecotect software to simulate the illumination condition of missing values in the data set, and uses the homogeneous mean interpolation method to complement the data; the specific operation is as follows: the method comprises the steps of importing weather data of a place where a photovoltaic test base of a saline-alkali soil is located into software, establishing a 3D simulation model of a photovoltaic panel array, simulating shading conditions of each measured position, fusing a mean interpolation method and a clustering interpolation method, processing missing values in a data set by using the same-class mean interpolation method which is more in line with actual conditions, taking a mean value or directly interpolating photosynthetic effective radiation corresponding to a position point of the same shading condition of a point to be processed, and reducing accumulated errors of subsequent experiments.
Then, the principle of integral superposition is applied to the processed data set to convert the instantaneous quantity into the accumulated quantity so as to meet the project requirement, and the specific operation method is as follows:
as shown in formula (1), the photosynthetic effective radiation data is multiplied by 0.28 square meter (average plant occupation area), and then the photosynthetic effective accumulation amount (effective light quantity) received at the position can be obtained by integrating the time period from 0 to t, wherein the formula is as follows:
E=∫ 0 t PAR*0.28（1）；
wherein: e is the accumulated effective light amount in the period of 0 to t,PARis photosynthetically active radiation; firstly taking t as 15 minutes, calculating the effective light quantity in each 15 minutes, superposing the effective light quantity in one day to obtain the energy obtained after the position is subjected to continuous change of the illumination shading condition, and processing the obtained daily effective light quantity data set into a sample set of a fixed time window.
The effective light quantity data set is formed by dividing time series data into a plurality of continuous time windows, each window comprises n+m continuous historical light quantity data points, n is a historical time step used for training, each time step comprises three input features, namely effective light quantity on the same day, highest air temperature and average air humidity, m is a time span of future data to be predicted, one day light quantity of continuous n days is used as an input of a model, accumulated effective light quantity of m days is used as an output of the model, and the data set is represented by 7:2: the scale of 1 is divided into a training set, a verification set and a test set.
The specific operation is as follows: firstly, loading data, converting a data format into an array, carrying out maximum and minimum normalization on the data, scaling the data to a range from 0 to 1, eliminating the influence caused by different characteristic scale differences, setting a time window step length n+m and a window moving step length 1, slicing the data into samples, wherein the number of the samples is equal to the data quantity- (n+m), and thus, maximally utilizing an original data set;
in order to improve the generalization performance of the model, the samples are randomly disturbed, and the sample set is subjected to a step of 7:2:1 into training set, verification set and test set, and finally selecting effective light quantity for n continuous daysX=[E 1 ,…,E n ]As an input sequence of the model,h=[E n+1 ,…,E n+m ]as an output sequence of the model, fig. 4 and 5 show.
Step three: constructing a loss function with priori knowledge constraint by using expert experience, and guiding a model training process, wherein the specific steps are as follows:
determining a value range of an output value of a model and a change rate range of the output value, namely determining a priori constraint rules, wherein in the specific embodiment of the invention, two constraint rules are respectively shown in the following formulas (2) and (3):
constraint rule 1: y is∈[34641.6,440714.4]（2）；
Constraint rule 2: rate of change= | [ y (t+1) -y (t)]/y(t)|∈[0.05%,567%]（3）；
Wherein y (t) is the predicted result of the effective light quantity model on the t th day after normalization;
the loss function is customized, and the model loss is selected from the average absolute error functionMAEThe calculation formula is as follows:
（4）；
in the method, in the process of the invention,MAErepresents the average of the absolute error between the predicted value and the observed value,y i for the predicted values at different times for each sample,y ture is the true value of its corresponding instant.
Programming the constraint rule, calculating an upper limit difference value between a predicted value and the constraint rule 1, loading a maximum instruction by using a tensorflow function library provided by python, returning a maximum value between the difference value and 0 as an upper limit loss of the constraint rule 1, and comparing the maximum value with the maximum value by using a lower limit loss calculation mode and a change rate upper limit loss calculation mode of the constraint rule 2, wherein the upper limit loss and the lower limit loss calculation mode are similar, namely, the predicted value/change rate and the limit boundary are different and the maximum value is obtained by comparing the difference value with the 0 value; and finally, adding the three loss parts to obtain the overall loss of the model, wherein the overall loss is shown in the following formula (5):
Loss=l MAE (y i ,y ture )+l (1) (y i )+l (2) (y)（5）；
wherein,l (1) (y i ) Is based on constraint rule 1 i.e. the loss resulting from the output range constraint of y,l (2) (y) Is a loss generated under the constraint of constraint rule 2, namely, the limiting condition of the change rate of the output at two adjacent moments of y.
Step four: an effective light quantity data design algorithm for one position is used for constructing an encoder-decoder model based on an LSTM layer (PK-LSTM) with a Priori Knowledge (PK) loss function, training the mapping relation of a light quantity sequence to a sequence, realizing the medium-long term prediction of the effective light quantity based on historical data, and respectively establishing an effective light quantity prediction model by applying the algorithm to other positions, so that the condition of receiving the light quantity of the whole saline-alkali soil in the future for n days under the same condition can be mastered.
As shown in fig. 2, the built encoder-decoder model includes 3 long-short-term memory network layers and 1 full-connection layer, and the specific steps of prediction by the model are as follows:
training data samples are input into the encoder in sequence, and in the process, the hidden state of the current moment is utilized to forget, memorize and learn by using the PK-LSTM layerh t Is hidden from the last momenth t-1 And current inputx t Co-influence, i.e. h t =f(h t-1 ,X t )，fFor the selected activation function, here the ReLu function, then the input sequence is encoded as a semantic vector c containing context features, which is the state of the encoder encoding stage for the final hidden layerh n Obtained by linear transformation q, i.e. c=q (h n ) The method comprises the steps of carrying out a first treatment on the surface of the The semantic vector c is used as the initial state of the decoding stage of the decoder, and the state of the last hidden layerh n As initial input to the decoding stage of a decodery 0 。
Taking c as the input of each neuron of the decoder, the hidden state at time th t Is hidden from the last momenth t-1 And output at the last timey t-1 And semantic sequence c influence, i.e. h t =f(h t-1 ,y t-1 C), after decoding different features of the semantic sequence c, reconstructing a label corresponding to the sample by a decoder, and finally realizing sequence-to-sequence prediction through a full connection layer.
After the data set is input into the model, forgetting, memorizing and learning are carried out on the input sequence X through a PK-LSTM layer, a semantic vector c containing the context relation characteristic is obtained through an encoder, a predicted sequence is obtained through a decoder reconstructing the semantic sequence c, the model is trained through an objective function with priori constraint guided by expert experience, parameters of the model are adjusted according to the performance of the model on a verification set so as to obtain the best training effect, and the training effect comprises the learning rate lr, training batch epoch, iteration times bs, the best neuron number units and the like of an Adam learner. The model is stored and tested to obtain the medium-long term prediction result of the effective light quantity, and the time step of the training set label is adjusted to obtain the predicted light quantity of any day in the future to achieve the purpose of prediction, as shown in fig. 6.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and not for limiting the same; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some or all of the technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit of the invention.

Claims (1)

1. A modeling prediction method for plant light receiving quantity under a saline-alkali soil photovoltaic panel is characterized by comprising the following steps of: the method comprises the following modeling prediction steps:
step one: collecting photosynthetic effective radiation, air temperature and humidity data in a target area by adopting instrument equipment, continuously collecting annual photosynthetic effective radiation instantaneous data at a plurality of different positions between two adjacent rows of photovoltaic panels paved in a saline-alkali soil at preset time intervals, and establishing a data set;
step two: the illumination condition of the missing value in the data set is simulated and analyzed, the missing value is supplemented by using a homogeneous mean interpolation method, the instantaneous quantity of photosynthetic effective radiation is converted into effective photosynthetic accumulation quantity by adopting an integral mode, and a model is built, wherein the specific method comprises the following steps:
introducing weather data of a place where a saline-alkali soil photovoltaic test base is located into Ecotect software, establishing a 3D simulation model of a photovoltaic panel array, simulating shading conditions of each measured position, and obtaining simulation data of photosynthetic effective radiation;
multiplying photosynthetic effective radiation data by the average plant occupied area, and integrating the time period from 0 to t to obtain the photosynthetic effective accumulation amount received in the time period from 0 to t of the position;
superposing the effective light quantity in one day to obtain the energy obtained after the position is subjected to continuous change of illumination shading conditions, loading data, converting the data format into an array, carrying out maximum and minimum normalization on the data, scaling the data to a range from 0 to 1, setting a time window step length n+m, a window moving step length 1, slicing the data into samples, and calculating the formula:
number of samples = data amount- (n+m);
wherein n is a historical time step for training, each time step comprises three input features, namely effective light quantity, highest air temperature and average air humidity of the same day, m is a time span of future data to be predicted, and the unit is days;
in order to improve the generalization performance of the model, samples are randomly disturbed, and a sample set is subjected to 7:2:1 is divided into a training set, a verification set and a test set;
selecting effective light quantity for n consecutive daysX=[E 1 ,…,E n ]As an input sequence of the model,h=[E n+1 ,…,E n+m ]as an output sequence of the model;
step three: constructing a loss function with priori knowledge constraint based on expert experience, guiding a model training process, providing useful priori information under the condition of limited data or noise of the data, and limiting the predicted output of the model to a specified range;
the specific method for constructing the loss function with priori knowledge constraint is as follows:
determining a priori constraint rule by determining a value range of an output value of a model and a change rate range of the output value, and defining two constraint rules, wherein the two constraint rules are specifically as follows:
constraint rule 1: y is∈[34641.6,440714.4]；
Constraint rule 2: rate of change= | [ y (t+1) -y (t)]/y(t)|∈[0.05%,567%]；
Wherein y (t) is the predicted result of the effective light quantity model on the t th day after normalization;
the built custom loss function comprises: average absolute error functionMAEA loss caused by constraint rule 1, a loss caused by constraint rule 2, wherein the average absolute error function is selectedMAEThe calculation formula of (2) is as follows:
；
in the method, in the process of the invention,MAErepresents the average of the absolute error between the predicted value and the observed value,y i for the predicted values at different times for each sample,y ture for the true value of its corresponding instant in time,mfor a predicted time step;
step four: an encoder-decoder prediction model is constructed on the basis of a long-term memory neural network layer with a loss function constrained by priori knowledge aiming at an effective light quantity prediction data design algorithm of a position in a target area, and a mapping relation from a historical effective light quantity sequence to a future effective light quantity sequence is trained to realize effective light quantity prediction based on historical data;
applying a design algorithm to other positions to respectively establish an effective light quantity prediction model, and obtaining the future light quantity receiving condition of the whole saline-alkali soil under the same condition based on the prediction model;
the specific method for forecasting by the effective light quantity forecasting model comprises the following steps:
the built encoder-decoder model comprises 3 long-period and short-period memory network layers and 1 full-connection layer, the data samples are trained through the model, the data samples are input into the encoder in sequence, the loss function with priori knowledge constraint and the neural network layer with long-period memory are fused for forgetting, memorizing and learning, and the hidden state at the current moment is adoptedh t Is hidden from the last momenth t-1 And current inputx t The common influence satisfies the expression:
h t =f(h t-1 ,X t )；
wherein,ffor the selected activation function, in particular the ReLu function;
the input sequence is then compiled into a semantic vector c containing contextual characteristics, the semantic vector c being defined in particular by the state of the final hidden layerh n The calculation formula is as follows:
c=q(h n )；
the semantic vector c is used as the initial state of the decoding stage of the decoder, and the mostThe state of the latter hidden layerh n As initial input to the decoding stage of a decodery 0 ；
Using the semantic vector c as the input of each neuron of the decoder to set the hidden state at the time th t Is hidden from the last momenth t-1 And output at the last timey t-1 And semantic sequence c effects, expressed as:
h t =f(h t-1 ,y t-1 ,c)；
after decoding different features of the semantic vector c, the decoder reconstructs the label corresponding to the sample, and finally, the sequence-to-sequence prediction is realized through the full connection layer.
CN202311574449.4A 2023-11-23 2023-11-23 Modeling prediction method for plant light receiving quantity under saline-alkali soil photovoltaic panel Active CN117272851B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202311574449.4A CN117272851B (en) 2023-11-23 2023-11-23 Modeling prediction method for plant light receiving quantity under saline-alkali soil photovoltaic panel

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202311574449.4A CN117272851B (en) 2023-11-23 2023-11-23 Modeling prediction method for plant light receiving quantity under saline-alkali soil photovoltaic panel

Publications (2)

Publication Number Publication Date
CN117272851A CN117272851A (en) 2023-12-22
CN117272851B true CN117272851B (en) 2024-02-02

Family

ID=89220049

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202311574449.4A Active CN117272851B (en) 2023-11-23 2023-11-23 Modeling prediction method for plant light receiving quantity under saline-alkali soil photovoltaic panel

Country Status (1)

Country Link
CN (1) CN117272851B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117591920A (en) * 2024-01-17 2024-02-23 北京科百宏业科技有限公司 Method and device for supplementing historical data, electronic equipment and readable storage medium

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110516844A (en) * 2019-07-25 2019-11-29 太原理工大学 Multivariable based on EMD-PCA-LSTM inputs photovoltaic power forecasting method
CN113722375A (en) * 2021-08-10 2021-11-30 国网浙江省电力有限公司电力科学研究院 Double-layer hybrid photovoltaic prediction method based on multi-algorithm optimization and computer equipment
CN115796004A (en) * 2022-11-04 2023-03-14 华北电力大学 Photovoltaic power station ultra-short term power intelligent prediction method based on SLSTM and MLSTNet models

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150012258A1 (en) * 2013-07-08 2015-01-08 Holden R. Caine System and method for modeling and characterizing of photovoltaic power systems
CN111626506B (en) * 2020-05-27 2022-08-26 华北电力大学 Regional photovoltaic power probability prediction method based on federal learning and cooperative regulation and control system thereof
CN113128793A (en) * 2021-05-19 2021-07-16 中国南方电网有限责任公司 Photovoltaic power combination prediction method and system based on multi-source data fusion
CA3178364A1 (en) * 2021-10-04 2023-04-04 Royal Bank Of Canada System and method for machine learning architecture for multi-task learning with dynamic neural networks

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110516844A (en) * 2019-07-25 2019-11-29 太原理工大学 Multivariable based on EMD-PCA-LSTM inputs photovoltaic power forecasting method
CN113722375A (en) * 2021-08-10 2021-11-30 国网浙江省电力有限公司电力科学研究院 Double-layer hybrid photovoltaic prediction method based on multi-algorithm optimization and computer equipment
CN115796004A (en) * 2022-11-04 2023-03-14 华北电力大学 Photovoltaic power station ultra-short term power intelligent prediction method based on SLSTM and MLSTNet models

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
An Experimental Study of LSTM Encoder-Decoder Model for Text Simplification;Tong Wang 等;《arXiv》;1-8 *
基于先验知识的长短记忆RBF网络结构;韩丽 等;《华北电力大学学报(自然科学版)》;第35卷(第5期);78-83 *
基于深度学习的LSTM光伏预测;崔承刚 等;《上海电力学院学报》;第35卷(第6期);544-552, 579 *

Also Published As

Publication number Publication date
CN117272851A (en) 2023-12-22

Similar Documents

Publication Publication Date Title
CN110084367B (en) Soil moisture content prediction method based on LSTM deep learning model
Jallal et al. A novel deep neural network based on randomly occurring distributed delayed PSO algorithm for monitoring the energy produced by four dual-axis solar trackers
CN117272851B (en) Modeling prediction method for plant light receiving quantity under saline-alkali soil photovoltaic panel
CN112036658B (en) Wind power real-time prediction method based on forgetting long-term and short-term memory with difference
CN116630122B (en) Lake ecological hydraulic regulation and control method and system based on hydrologic-ecological response relation
CN114626512A (en) High-temperature disaster forecasting method based on directed graph neural network
CN102183802B (en) Short-term climate forecast method based on Kalman filtering and evolution modeling
Kalogirou Artificial neural networks and genetic algorithms in energy applications in buildings
CN114492922A (en) Medium-and-long-term power generation capacity prediction method
Li et al. A novel multichannel long short-term memory method with time series for soil temperature modeling
CN114004430B (en) Wind speed forecasting method and system
CN115222163A (en) Multi-factor medium-long term real-time forecasting method and system for harbor basin inlet waves and application
CN117031582B (en) Ozone hour concentration forecasting method based on recursive space-time learning and simulation monitoring fusion
CN117909888B (en) Intelligent artificial intelligence climate prediction method
CN115456245A (en) Prediction method for dissolved oxygen in tidal river network area
CN116013426A (en) Site ozone concentration prediction method with high space-time resolution
Wu et al. Two-stage decomposition and temporal fusion transformers for interpretable wind speed forecasting
Wu et al. Interpretable wind speed forecasting with meteorological feature exploring and two-stage decomposition
Bamisile et al. Long-term prediction of solar radiation using XGboost, LSTM, and machine learning algorithms
CN117114438A (en) Building area energy system cold and hot load data driving prediction method with flexibility and interpretability
CN116470486A (en) Ultra-short-term photovoltaic power generation power prediction method, device and storage medium
CN115907204A (en) Forest transpiration water consumption prediction method for optimizing BP neural network by sparrow search algorithm
Martin et al. Neural network development for the forecasting of upper atmosphere parameter distributions
CN114418167A (en) Plateau mountain multi-meteorological new energy power generation prediction optimization method
CN113077110A (en) GRU-based harmonic residual segmented tide level prediction method

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant