LSTM-based cross-seasonal plant water resource demand prediction method and system

Abstract

The application provides a LSTM-based cross-season plant water resource demand prediction method and system, and relates to the technical field of data processing, which comprises the following steps: obtaining historical data of a cross-season plant growth area; normalizing the time index of the historical data to obtain a time index normalization coefficient; decomposing historical water resource demand through Chebyshev basis functions embedded with the time normalization coefficient to obtain seasonal components and non-seasonal components describing the historical water resource demand; combining historical environmental data, the seasonal components and the non-seasonal components to construct water resource description features; inputting the water resource description features into a dynamic gated LSTM network to obtain predicted non-seasonal components; determining predicted seasonal components in combination with Chebyshev spectrum coefficients; and determining the predicted water resource demand of the cross-season plant growth area in combination with the predicted non-seasonal components and the predicted seasonal components. The water resource demand prediction accuracy of the cross-season key turning point is significantly improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for predicting cross-seasonal plant water resource demand based on LSTM. Background Technology

[0002] LSTM (Long Short-Term Memory) is a special type of recurrent neural network (RNN) that addresses the vanishing and exploding gradient problems encountered by standard RNNs in processing long-sequence data. LSTM uses a gating mechanism to control the flow of information, enabling the network to retain historical information over a longer period, thus effectively capturing long-term dependencies in prediction tasks. Transseasonal plants are those that grow in different seasons and require different environmental conditions. Their water resource needs are influenced by climate change, seasonal precipitation, and temperature variations, exhibiting transseasonal characteristics. While their water resource requirements are typically periodic, they can also be affected by non-seasonal factors such as extreme weather.

[0003] Cross-seasonal water demand forecasting helps agricultural managers plan and allocate water resources more accurately across different seasons, avoiding resource waste or shortages caused by seasonal variations. Accurate water demand forecasting can improve irrigation efficiency, reduce water consumption, ensure adequate water supply to plants at different stages, and promote healthy growth. Furthermore, this forecasting is crucial for addressing the impacts of climate change and improving the sustainability and efficiency of agricultural production.

[0004] However, existing technologies struggle to accurately decompose the seasonal and non-seasonal components in complex historical data when forecasting cross-seasonal plant water demand. In particular, they fail to accurately capture long-term dependent patterns in water demand, leading to significant error accumulation in cross-seasonal plant water demand forecasting. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the purpose of this invention is to provide an LSTM-based method for predicting cross-seasonal plant water resource demand, which can solve the technical problem that the prior art is unable to accurately decompose the seasonal and non-seasonal components in the non-stationary sequence of complex historical data when predicting cross-seasonal plant water resource demand, especially the difficulty in accurately capturing the long-term dependent water resource demand pattern, which leads to significant error accumulation in cross-seasonal plant water resource demand prediction.

[0006] A first aspect of this invention proposes a method for predicting cross-seasonal plant water resource demand based on LSTM, comprising:

[0007] S1: Obtain historical data of cross-seasonal plant growth areas, wherein the historical data includes historical environmental data and historical water resource requirements;

[0008] S2: Normalize the time index of the historical data to obtain the time index normalization coefficient;

[0009] S3: Decompose the historical water resource demand by using a Chebyshev basis function with embedded time normalization coefficients to obtain seasonal and non-seasonal components describing the historical water resource demand;

[0010] S4: Construct water resource description features by combining the historical environmental data, the seasonal component, and the non-seasonal component;

[0011] S5: Input the water resource description features into a dynamically gated LSTM network to obtain the predicted non-seasonal components, wherein the dynamically gated LSTM network includes cell states with respect to the first type of Chebyshev polynomials;

[0012] S6: Determine the predicted seasonal components by combining Chebyshev spectral coefficients;

[0013] S7: Combine the predicted non-seasonal component and the predicted seasonal component to determine the predicted water resource demand of the cross-seasonal plant growth zone.

[0014] A second aspect of the present invention provides a cross-seasonal plant water resource demand prediction system based on LSTM, comprising: a processor and a memory;

[0015] The memory stores a program or instructions that can run on the processor, and when the program or instructions are executed by the processor, they implement the steps of the LSTM-based method for predicting cross-seasonal plant water demand as described in the first aspect.

[0016] A third aspect of the present invention provides a readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the steps of the LSTM-based method for predicting cross-seasonal plant water resource demand as described in the first aspect.

[0017] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0018] In this embodiment of the invention, the complex seasonal fluctuations and non-stationary trend components of historical water resource demand are accurately decomposed by combining Chebyshev basis functions with normalized time indexes. Then, environmental data is integrated to construct a comprehensive water resource descriptive feature input to a dynamically gated LSTM network. This network utilizes Chebyshev polynomials to enhance cell states, significantly improving the ability to model long-term dependencies and effectively capturing non-seasonal dynamic patterns. Simultaneously, seasonal components are stably extrapolated based on Chebyshev spectral coefficients. Finally, prediction is achieved through component reconstruction, accurately separating non-stationary historical data and overcoming the defects of fuzzy decomposition. The dynamically gated LSTM network structure solves the error accumulation problem under long-term dependencies, and the synergistic effect of multi-source features strengthens the correlation modeling between environmental factors and water demand patterns, significantly improving the accuracy of water resource demand prediction at key seasonal turning points. Attached Figure Description

[0019] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0020] Figure 1 This is a flowchart illustrating a cross-seasonal plant water resource demand prediction method based on LSTM provided in an embodiment of the present invention.

[0021] Figure 2 This is a schematic diagram of the structure of a cross-seasonal plant water resource demand prediction system based on LSTM provided in an embodiment of the present invention. Detailed Implementation

[0022] To enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. It should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0023] The following description, in conjunction with the accompanying drawings, details the LSTM-based method for predicting cross-seasonal plant water resource demand provided by the present invention through specific embodiments and application scenarios.

[0024] Reference manual attached Figure 1 The diagram illustrates a flowchart of a cross-seasonal plant water resource demand prediction method based on LSTM provided in an embodiment of the present invention.

[0025] This invention provides a method for predicting cross-seasonal plant water resource demand based on LSTM, which may include the following steps:

[0026] S1: Obtain historical data of cross-seasonal plant growth zones.

[0027] Historical data includes historical environmental data and historical water resource requirements. Historical environmental data includes temperature and humidity data for cross-seasonal plant growth zones.

[0028] Specifically, the cross-seasonal plant growth zone refers to the area where cross-seasonal plants grow, requiring water resource demand assessment. By collecting historical data from these zones, a reliable historical basis for forecasting can be provided, helping to accurately identify plant growth patterns and trends in water resource demand, thereby improving the accuracy and reliability of forecasts.

[0029] S2: Normalize the time index of historical data to obtain the time index normalization coefficient.

[0030] The time index refers to a recorded point in time or time period in the data, used to identify the specific time when the data occurred. The time index corresponds to each point in time (such as date, hour, etc.) in historical data, and it is used to track changes in plant water resource requirements and environmental data over time. The time index normalization coefficient is the coefficient obtained after normalizing the original time index. By normalizing the time index, the scale differences of time data can be effectively eliminated, allowing data from different time periods to be processed under a unified standard. This avoids interference with prediction results from time spans that are too large or too small, thereby improving prediction stability and accuracy.

[0031] In one possible implementation, the time index normalization coefficient is specifically the quotient of the numerator and the denominator, wherein the numerator is specifically the difference between the time index and the sum of the maximum and minimum time indices, and the denominator is specifically the difference between the maximum and minimum time indices.

[0032] It's important to note that normalizing the time index in historical data standardizes data from different points in time, placing them within the same range. By calculating the ratio of the time index to the difference between the maximum and minimum time indices, the impact of time span differences on the data is eliminated, allowing data from different time periods to be processed under a uniform standard. This normalization method effectively avoids prediction bias caused by time scale differences, improves the stability and accuracy of predictions, ensures the rationality of time data, and provides more consistent and accurate input for subsequent analysis. It helps reduce errors caused by time variations, thereby improving the reliability of prediction results.

[0033] S3: By decomposing historical water demand using Chebyshev basis functions with embedded time-normalized coefficients, we obtain the seasonal and non-seasonal components describing historical water demand.

[0034] Chebyshev basis functions are a special type of mathematical function commonly used in numerical analysis and signal processing. They can efficiently approximate the variation patterns of complex functions. By embedding time normalization coefficients into Chebyshev basis functions, historical water demand data can be effectively decomposed. Seasonal components refer to the periodic variations in water demand data, which are typically influenced by seasonal climatic factors (such as temperature and precipitation). For example, water demand is higher in summer and lower in winter; these periodic fluctuations belong to the seasonal component. Non-seasonal components refer to the parts of water demand data that do not change with the seasons, typically representing long-term trends or the influence of other non-periodic factors, such as long-term trends in plant growth or the long-term effects of climate change.

[0035] By decomposing historical water demand data using Chebyshev basis functions, it is accurately broken down into seasonal and non-seasonal components, allowing for the separate handling of periodic and non-periodic variations. This helps to better capture the complex patterns of water demand, reduce unnecessary errors, and provide clearer and more accurate data input for subsequent forecasting.

[0036] In one possible implementation, S3 specifically includes:

[0037] S301: Recursively generate Chebyshev basis functions based on time normalization coefficients, wherein the Chebyshev basis functions include zero-order Chebyshev basis functions, first-order Chebyshev basis functions, and higher-order Chebyshev basis functions.

[0038] The specific formula for Chebyshev basis functions is as follows:

[0039] ;

[0040] in, This indicates the order of the Chebyshev basis functions. , and Let represent the zeroth-order Chebyshev basis function, the first-order Chebyshev basis function, and the kth-order Chebyshev basis function with respect to the time index t, respectively. and Let them represent the (k-1)th order Chebyshev basis functions and the (k-2)th order Chebyshev basis functions, respectively. and These represent the maximum and minimum time indices, respectively. This represents the time normalization coefficient.

[0041] It should be noted that this process recursively generates Chebyshev basis functions based on time normalization coefficients, progressively constructing higher-order basis functions starting from order zero. Each order of Chebyshev basis function is generated according to the recursive relationship and normalized time index, used to describe different periodic patterns in the time series. The normalized time index ensures that data from different time ranges can be processed at the same scale, making the generation of Chebyshev basis functions adaptable to data with various time spans. By recursively generating these basis functions, the periodic changes and trends in historical water resource demand can be accurately captured. The advantage of this method is that Chebyshev basis functions can efficiently approximate the periodic characteristics of complex data, thereby decomposing the seasonal and non-seasonal components in the data, which helps improve the accuracy and predictive ability of the model, especially when dealing with nonlinear and highly complex data.

[0042] S302: Project historical water resource demand onto Chebyshev basis functions of different orders to obtain Chebyshev spectral coefficients describing the intensity of periodic modes of different orders.

[0043] The intensity of different order periodic patterns refers to the intensity of different periodic change patterns in historical water resource demand. Each order of Chebyshev basis function corresponds to a different frequency or periodic characteristic. Each order (e.g., zeroth, first, second, etc.) of Chebyshev basis function represents a different periodic pattern, and they describe the fluctuation or periodic change of data through different frequencies.

[0044] The specific formula for calculating the Chebyshev spectral coefficients is as follows:

[0045] ;

[0046] in, Represents the k-th order Chebyshev spectral coefficients. This represents the total number of historical time points. This represents the historical water resource demand at time point t. Let represent the value of the k-th order Chebyshev basis function at time t.

[0047] The total number of time points is 365 observation time points per year.

[0048] S303: Combining the truncation order, the seasonal components are obtained by weighted summation of the Chebyshev spectral coefficients and Chebyshev basis functions of each order.

[0049] ;

[0050] in, This represents the seasonal component of historical water demand with respect to time index t. The truncation order represents the complexity of the constrained seasonal components of the Chebyshev basis functions.

[0051] It should be noted that those skilled in the art can set the cutoff order according to actual needs, and this invention does not limit it. Optionally, this value can be set to 2.

[0052] Specifically, this process calculates the Chebyshev spectral coefficients for each order by projecting historical water resource demand onto Chebyshev basis functions of different orders. These coefficients reflect the intensity of periodic patterns in the data. By weighted summing of the spectral coefficients of different orders with the Chebyshev basis functions, the seasonal component of water resource demand can be obtained, accurately capturing periodic fluctuations in the data. The use of a truncation order limits the complexity of the seasonal component, avoiding overfitting and ensuring the simplicity and effectiveness of the model. This method can effectively extract periodic features from the data, helping to more accurately predict seasonal changes in water resource demand, thereby improving the stability and accuracy of predictions.

[0053] In one possible implementation, the method for determining the truncation order specifically includes:

[0054] S3031: Calculate the square of the modulus of each order of Chebyshev spectral coefficient to describe the energy of the Chebyshev spectral coefficient.

[0055] It should be noted that, based on Passevar's theorem, the total energy of a signal under a discrete orthogonal basis is equal to the sum of the squares of the magnitudes of its spectral coefficients.

[0056] S3032: Calculate the proportion of Chebyshev spectral coefficient energy to the total energy of Chebyshev spectral coefficients under different candidate truncation orders.

[0057] S3033: Determine the comprehensive score for each candidate cutoff order based on the proportional coefficient.

[0058] The formula for calculating the overall score is as follows:

[0059] ;

[0060] ;

[0061] ;

[0062] in, The square of the modulus of the k-th order Chebyshev spectral coefficient is represented by the energy of the k-th order Chebyshev spectral coefficient. Indicates the candidate cutoff order. This represents the scaling factor corresponding to the candidate cutoff order j. This represents the overall score corresponding to the candidate truncation order j.

[0063] S3034: Output the candidate cutoff order corresponding to the highest comprehensive score as the cutoff order.

[0064] It should be noted that the energy (squared modulus) of the Chebyshev spectral coefficients at each order is calculated, and the contribution of the spectral coefficients to the total signal energy is evaluated based on Passevar's theorem. Next, the proportion of the energy of the Chebyshev spectral coefficients to the total energy is calculated for different candidate truncation orders, and a comprehensive score for each candidate truncation order is determined based on this proportion. The comprehensive score evaluates the optimal truncation order by considering both the energy proportion and the contribution of the order to the total energy, and finally, the candidate truncation order with the highest comprehensive score is selected as the output. This method can effectively identify the most representative periodic components, avoid overfitting or information loss, ensure the accuracy and simplicity of the decomposition, thereby improving the model's ability to capture seasonal components and ensuring the stability and reliability of the prediction.

[0065] S304: Remove the seasonal component from historical water resource demand to obtain the non-seasonal component.

[0066] The specific formula for calculating the non-seasonal component is as follows:

[0067] ;

[0068] in, This represents the non-seasonal component at different time points t.

[0069] Specifically, this step decomposes historical water demand data using Chebyshev basis functions. First, after normalizing the time index, basis functions of different orders are recursively generated using Chebyshev basis functions. Next, the historical water demand data is projected onto these basis functions, and the Chebyshev spectral coefficients for each order are calculated, representing the intensity of different periodic patterns. By weighted summing of the Chebyshev spectral coefficients of different orders with the corresponding basis functions, the seasonal component is obtained, representing the periodic variation in the data. The non-seasonal component is obtained by removing the seasonal component from the historical data, representing the influence of long-term trends and other non-periodic factors. This process accurately decomposes the seasonal and non-seasonal components of historical water demand, allowing different variation patterns to be handled separately, thereby reducing errors in the data, enhancing the model's adaptability to complex water demand patterns, providing a more accurate basis for subsequent forecasts, and improving the stability and reliability of forecasts.

[0070] S4: Construct water resource descriptive features by combining historical environmental data, seasonal components, and non-seasonal components.

[0071] Among them, the water resource descriptive features are a set of features extracted by combining historical environmental data (such as temperature and humidity), seasonal components, and non-seasonal components to describe water resource demand. These water resource descriptive features help capture the multiple factors affecting water resource demand, thereby improving the accuracy of forecasts and adaptability to complex change patterns.

[0072] In one possible implementation, the water resource description features are specifically a feature vector obtained by combining historical environmental data, seasonal components, and non-seasonal components.

[0073] S5: Input the water resource description features into a dynamically gated LSTM network to obtain the predicted non-seasonal components.

[0074] The dynamically gated LSTM network includes cell states with respect to Chebyshev polynomials of the first class.

[0075] Dynamically gated LSTM networks are a special type of neural network architecture primarily used to handle long-term dependencies in time series data. Dynamic gating refers to the LSTM network's ability to dynamically adjust its processing method based on different input data, optimizing the modeling of time series. This network can effectively identify and predict long-term dependencies. Chebyshev polynomials of the first kind are a class of polynomials with special mathematical properties, widely used in numerical analysis and signal processing. The definition of Chebyshev polynomials of the first kind is recursive, and they possess a minimum maximum error, making them well-suited for approximating complex functions. Cell states are a core mechanism in LSTM networks, used to maintain and pass crucial information between time steps, thereby ensuring the network can learn and retain information about long-term dependencies.

[0076] By inputting water resource description features into a dynamically gated LSTM network and combining it with Chebyshev multinomial enhancement of cell states, long-term dependencies can be better captured, and the non-seasonal components of water resource demand can be effectively predicted, thereby improving the model's predictive ability in complex and non-periodic changes.

[0077] In one possible implementation, S5 specifically includes:

[0078] S501: Determine cell state by combining the first-type Chebyshev polynomial.

[0079] The specific formula for calculating cell state is as follows:

[0080] ;

[0081] ;

[0082] ;

[0083] ;

[0084] ;

[0085] ;

[0086] in, , , , and These represent the cell state, forget gate, input gate, candidate state, and modulation term related to time point t, respectively. This represents the cell state relative to time point t-1. This represents element-wise multiplication. This represents the sigmoid activation function. Let M represent the hyperbolic tangent activation function, and M represent the cutoff order corresponding to the highest overall score. Denotes a Chebyshev polynomial of the first kind of order k. and These represent the weight parameters and bias parameters of the forget gate, respectively. This represents the hidden state corresponding to time point t-1. This represents the water resource description characteristics corresponding to time point t. and These represent the weight parameters and bias parameters of the output gate, respectively. and These represent the weight parameters and bias parameters of the cell state, respectively. This represents a second-order Chebyshev polynomial. This represents the non-seasonal component residual corresponding to time point t-1. and These represent the non-seasonal components corresponding to time points t-1 and t-2, respectively.

[0087] It should be noted that the calculation of cell states involves the dynamic adjustment of forget gates, input gates, candidate states, and modulation terms. The forget gate determines which information should be forgotten, the input gate controls the introduction of new information, candidate states process the input data using a hyperbolic tangent activation function, and the modulation term regulates the cell states using the residuals of non-seasonal components and a second-order Chebyshev polynomial. This approach, combining Chebyshev polynomials, accurately enhances the memory of non-seasonal components in the data, especially when dealing with complex non-periodic changes, where cell states effectively retain key information. By progressively adjusting these gating mechanisms and states, the model can improve its ability to predict long-term trends, thereby enhancing the accuracy of water resource demand forecasting.

[0088] S502: Calculate the hidden state corresponding to the last time step in the sliding window based on the cell state iteration.

[0089] The specific formula for calculating the hidden state is as follows:

[0090] ;

[0091] ;

[0092] in, This represents the hidden state corresponding to the last time step N. and These represent the cell state and output gate corresponding to the last time step N, respectively. This represents the hyperbolic tangent activation function. This represents the sigmoid activation function.

[0093] The sliding window describes the number of historical time steps used when predicting future water demand. Optionally, it can be set to 7 days or 31 days.

[0094] S503: Input the hidden state into the fully connected layer of the dynamically gated LSTM network to obtain the predicted non-seasonal component.

[0095] The specific formula for calculating the predicted non-seasonal component is as follows:

[0096] ;

[0097] in, This indicates that the non-seasonal component corresponding to time point t+1 is the predicted non-seasonal component. and These represent the weight parameters and bias parameters of the fully connected layer, respectively.

[0098] Specifically, this process inputs water resource descriptive features into a dynamically gated LSTM network and incorporates Chebyshev polynomials to enhance cell states, thereby better capturing long-term dependencies in the data. The LSTM network, through its cell state mechanism, transmits key information between time steps, ensuring the learning and retention of long-term patterns of change. Combined with Chebyshev polynomials, the network effectively enhances its ability to capture non-seasonal components (such as long-term trends or abrupt events). By calculating the hidden state corresponding to the last time step within the sliding window, the network can progressively adjust its output and make accurate predictions. Finally, the prediction results output the non-seasonal components through fully connected layers. This method can reduce the impact of long-term dependencies when dealing with complex, non-periodic water resource demand data, enhancing the accuracy and stability of predictions. By dynamically adjusting the network's processing method, its adaptability to different types of data patterns is optimized, improving the model's performance in future demand forecasting.

[0099] The dynamically gated LSTM (Long Short-Term Memory) network relies on cell states to handle long-term dependencies in time-series data. The LSTM network comprises the following main components: Forget Gate: Determines how much information from the previous time step should be discarded at each time step. Calculated using a sigmoid activation function, the output of the forget gate is multiplied element-wise with the cell state from the previous time step to determine which information will be "forgotten." Input Gate: Controls the input of new information. Calculated using a sigmoid activation function, the input gate determines how much new data should be added to the current cell state. Candidate States: Transform the input using the output of the input gate and the hidden state from the previous time step to generate new candidate states. These candidate states, along with the output of the input gate, update the cell state. Cell State: The core of the LSTM, the cell state transmits information within the network and determines how the model remembers past knowledge. The cell state is updated through a weighted forget gate and the output of the input gate, preserving information about long-term dependencies. Modulation Term: Further adjusts the cell state by combining the residuals of non-seasonal components with Chebyshev polynomials to strengthen the memory of non-seasonal components. Output gate: Determines when to output the cell state to the next time step and adjusts the final hidden state. Hidden state: The current state calculated through the output gate, which includes the output from the network to the next layer.

[0100] When training a dynamically gated LSTM network, historical water resource description features are first input into the network. The network dynamically adjusts cell states through its complex gating mechanism to capture long-term dependencies in the data. The goal during training is to update the network parameters (such as the weights and biases of the forget gate, input gate, and output gate) by optimizing the loss function. Using backpropagation, the network adjusts these parameters via gradient descent to minimize prediction error. Specifically, the network utilizes a sliding window technique, training at multiple time steps, with the output at each time step influencing the prediction at the next step. During training, the network continuously adjusts the gating mechanism and cell states to better capture long-term trends, especially in the case of non-seasonal variations. By progressively reducing errors, the training process enables the model to more accurately predict future non-seasonal components, thereby improving the stability and accuracy of water resource demand forecasting.

[0101] S6: Determine the predicted seasonal components by combining Chebyshev spectral coefficients.

[0102] Among them, the Chebyshev spectral coefficients are the projection coefficients of the historical water demand series onto the Chebyshev orthogonal basis functions. The predicted seasonal component is the seasonal fluctuation part at the future time t+1. By using historical Chebyshev spectral coefficients to predict future seasonal components, relying on the stability and minimum-maximum approximation properties of the Chebyshev spectral coefficients, the future seasonal components can be accurately predicted without repeatedly decomposing historical data, which improves prediction efficiency and ensures the consistency of the seasonal pattern.

[0103] In one possible implementation, S6 specifically includes:

[0104] S601: Calculate the normalized value at the predicted time point.

[0105] The formula for calculating the normalized value is as follows:

[0106] ;

[0107] in, and These represent the maximum and minimum time indices in the historical data's time index, respectively. Indicates the predicted time point The normalized value.

[0108] S602: Calculate the predicted seasonal components based on the normalized values ​​and Chebyshev spectral coefficients.

[0109] The specific formula for calculating the predicted seasonal component is as follows:

[0110] ;

[0111] in, This indicates the predicted seasonal component, and M represents the cutoff order corresponding to the highest overall score. Indicates the predicted time point The k-th order Chebyshev basis function values.

[0112] Specifically, this process predicts future seasonal components by leveraging the stability and minimum-maximum approximation properties of Chebyshev spectral coefficients. First, a normalized value is calculated for the prediction time point, ensuring that the data at that point can be compared with historical data on the same scale. Then, the predicted seasonal component is calculated using a weighted summation of the normalized value and the Chebyshev spectral coefficients. The Chebyshev spectral coefficients accurately capture seasonal fluctuations in historical data, and due to their minimum-maximum approximation properties, they can efficiently predict future seasonal components without re-decomposing historical data. This not only improves prediction efficiency but also ensures the consistency of seasonal patterns, making seasonal predictions more accurate and stable, and maintaining high accuracy even under long-term trend changes.

[0113] S7: Combine the predicted non-seasonal component and the predicted seasonal component to determine the predicted water resource demand for cross-seasonal plant growth zones.

[0114] In one possible implementation, forecasting water demand specifically involves forecasting the sum of the non-seasonal component and the seasonal component.

[0115] It should be noted that this process obtains the overall water demand forecast by adding the predicted non-seasonal and seasonal components. The non-seasonal component reflects long-term trends and sudden changes, while the seasonal component captures seasonally related cyclical fluctuations. By forecasting these two components separately and then synthesizing them, a more accurate and comprehensive description of the overall characteristics of water demand can be obtained, avoiding the shortcomings of focusing on only a single factor. This method can fully consider complex time series patterns while maintaining efficient computation and accurate forecasting, thus improving the overall predictive ability for future water demand.

[0116] In practical applications, this method combines a dynamically gated LSTM network and Chebyshev polynomials to accurately capture and predict cross-seasonal plant water demand. Specifically, firstly, by normalizing historical data and applying Chebyshev basis functions, water demand data is decomposed into seasonal and non-seasonal components. The non-seasonal component reflects long-term trends, while the seasonal component captures periodic fluctuations related to seasonal variations. By predicting these two components separately and then combining them, future water demand can be comprehensively and accurately predicted. Chebyshev polynomials provide efficient and accurate periodic pattern modeling, avoiding redundant historical data decomposition, thus improving computational efficiency and accuracy. This decomposition and prediction method ensures that the water demand model can stably and accurately predict future demand under the dual influence of long-term trends and seasonal fluctuations.

[0117] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0118] In this embodiment of the invention, the complex seasonal fluctuations and non-stationary trend components of historical water resource demand are accurately decomposed by combining Chebyshev basis functions with normalized time indexes. Then, environmental data is integrated to construct a comprehensive water resource descriptive feature input to a dynamically gated LSTM network. This network utilizes Chebyshev polynomials to enhance cell states, significantly improving the ability to model long-term dependencies and effectively capturing non-seasonal dynamic patterns. Simultaneously, seasonal components are stably extrapolated based on Chebyshev spectral coefficients. Finally, prediction is achieved through component reconstruction, accurately separating non-stationary historical data and overcoming the defects of fuzzy decomposition. The dynamically gated LSTM network structure solves the error accumulation problem under long-term dependencies, and the synergistic effect of multi-source features strengthens the correlation modeling between environmental factors and water demand patterns, significantly improving the accuracy of water resource demand prediction at key seasonal turning points.

[0119] Reference manual attached Figure 2 The diagram shows a schematic representation of a cross-seasonal plant water resource demand prediction system based on LSTM provided in an embodiment of the present invention.

[0120] This invention provides an LSTM-based cross-seasonal plant water demand prediction system 20, comprising: a processor 201 and a memory 202;

[0121] The memory 202 stores programs or instructions that can run on the processor 201. When the program or instructions are executed by the processor 201, they implement the steps of the above-described LSTM-based cross-seasonal plant water resource demand prediction method and achieve the same technical effect. To avoid repetition, the present invention will not elaborate further.

[0122] It should be understood that the processor 201 in this embodiment of the invention may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0123] It should also be understood that the memory 202 in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct memory bus RAM (DR RAM).

[0124] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0125] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0126] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0127] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0128] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0129] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0130] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0131] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0132] This invention provides a readable storage medium comprising: storing a program or instructions on the readable storage medium, wherein when the program or instructions are executed by a processor, the program or instructions implement the steps of the above-described LSTM-based method for predicting cross-seasonal plant water resource demand, and achieve the same technical effect. To avoid repetition, this invention will not elaborate further.

[0133] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting cross-seasonal plant water resource demand based on LSTM, characterized in that, include: S1: Obtain historical data of cross-seasonal plant growth areas, wherein the historical data includes historical environmental data and historical water resource requirements; S2: Normalize the time index of the historical data to obtain the time index normalization coefficient; S3: Decompose the historical water resource demand by using a Chebyshev basis function with embedded time normalization coefficients to obtain seasonal and non-seasonal components describing the historical water resource demand; S4: Construct water resource description features by combining the historical environmental data, the seasonal component, and the non-seasonal component; S5: Input the water resource description features into a dynamically gated LSTM network to obtain the predicted non-seasonal components, wherein the dynamically gated LSTM network includes cell states with respect to the first type of Chebyshev polynomials; S6: Determine the predicted seasonal components by combining Chebyshev spectral coefficients; S7: Combine the predicted non-seasonal component and the predicted seasonal component to determine the predicted water resource demand of the cross-seasonal plant growth zone.

2. The LSTM-based method for predicting cross-seasonal plant water resource demand according to claim 1, characterized in that, The time index normalization coefficient is specifically the quotient of the numerator and the denominator, wherein the numerator is specifically the difference between the time index and the sum of the maximum time index and the minimum time index, and the denominator is specifically the difference between the maximum time index and the minimum time index.

3. The LSTM-based method for predicting cross-seasonal plant water resource demand according to claim 1, characterized in that, S3 specifically includes: S301: Recursively generate the Chebyshev basis function based on the time normalization coefficient, wherein the Chebyshev basis function includes a zero-order Chebyshev basis function, a first-order Chebyshev basis function, and higher-order Chebyshev basis functions; S302: Project the historical water resource demand onto Chebyshev basis functions of different orders to obtain Chebyshev spectral coefficients describing the intensity of periodic patterns of different orders; S303: Combining the truncation order, the Chebyshev spectral coefficients of each order and the Chebyshev basis functions are weighted and summed to obtain the seasonal component; S304: Remove the seasonal component from the historical water resource demand to obtain the non-seasonal component.

4. The LSTM-based method for predicting cross-seasonal plant water resource demand according to claim 3, characterized in that, The method for determining the truncation order specifically includes: S3031: Calculate the square of the modulus of each order of Chebyshev spectral coefficients to describe the energy of the Chebyshev spectral coefficients; S3032: Calculate the proportion of Chebyshev spectral coefficient energy to the total energy of Chebyshev spectral coefficients under different candidate truncation orders; S3033: Determine the comprehensive score for each candidate cutoff order based on the aforementioned proportional coefficient; S3034: Output the candidate truncation order corresponding to the highest comprehensive score as the truncation order.

5. The LSTM-based method for predicting cross-seasonal plant water resource demand according to claim 1, characterized in that, The water resource description features are specifically a feature vector obtained by combining the historical environmental data, the seasonal component, and the non-seasonal component.

6. The LSTM-based method for predicting cross-seasonal plant water resource demand according to claim 1, characterized in that, S5 specifically includes: S501: Determine the cell state by combining the first type of Chebyshev polynomial; S502: Calculate the hidden state corresponding to the last time step in the sliding window based on the cell state iterative calculation; S503: Input the hidden state into the fully connected layer of the dynamically gated LSTM network to obtain the predicted non-seasonal component.

7. The LSTM-based method for predicting cross-seasonal plant water resource demand according to claim 1, characterized in that, S6 specifically includes: S601: Calculate the normalized value at the predicted time point; S602: Calculate the predicted seasonal component based on the normalized value and the Chebyshev spectral coefficient.

8. The LSTM-based method for predicting cross-seasonal plant water resource demand according to claim 1, characterized in that, The predicted water resource demand is specifically the sum of the predicted non-seasonal component and the predicted seasonal component.

9. A cross-seasonal plant water resource demand prediction system based on LSTM, characterized in that, include: Processor and memory; The memory stores programs or instructions that can run on the processor, which, when executed by the processor, implement the steps of the LSTM-based method for predicting cross-seasonal plant water demand as described in any one of claims 1 to 8.

10. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the LSTM-based method for predicting cross-seasonal plant water resource demand as described in any one of claims 1 to 8.

Images