Taro water and fertilizer coordination optimization method based on digitalized industrial internet collection

Abstract

The application discloses a taro water and fertilizer coordination optimization method based on digitalized industrial internet collection, and belongs to the technical field of intelligent agricultural planting management. Specifically, multi-dimensional data of a taro root area is collected, and an event-driven scale conversion value and an agricultural state correction value are constructed; a water-nutrient coupling deviation degree vector is constructed based on a water and fertilizer target parameter table and the agricultural state correction value; an improved genetic algorithm is constructed, including obtaining nitrogen-potassium gray correlation and phosphorus-potassium gray correlation by using sliding gray correlation, and then calculating a nitrogen phosphorus-potassium antagonistic intensity factor, fusing the deviation degree, the antagonistic factor and input penalty to construct a fitness function, adjusting the crossover and mutation probability based on the water deviation degree deterioration rate, and using chaos-reverse initialization in the initialization stage and using an elite reservation mechanism in the circulation stage; and the improved genetic algorithm is used to find an optimal water and fertilizer decision vector. The application fully considers the conditions of different stages during the development of taro, and can find an optimal water and fertilizer implementation scheme.

Description

Technical Field

[0001] This invention belongs to the field of smart agricultural planting management technology, specifically involving a method for synergistic optimization of water and fertilizer in taro based on digital IoT data collection. Background Technology

[0002] Water and fertilizer management are core agronomic measures affecting taro yield and quality. In traditional cultivation, farmers mostly follow the management method of "flood irrigation + experience-based fertilization": irrigation is carried out by furrow irrigation or bed flooding, keeping the furrows waterlogged throughout the growing season; fertilization is mainly based on basal fertilizer, with topdressing done haphazardly, and the nitrogen, phosphorus, and potassium ratios are crude. In recent years, integrated water and fertilizer technology has been widely used in greenhouse vegetables and fruit trees, but systematic research on root and tuber crops such as taro is still in its early stages. Therefore, it is essential to develop a water and fertilizer synergistic optimization technology system suitable for taro cultivation.

[0003] However, existing methods fail to distinguish between the impulsive step changes caused by irrigation and fertilization events and the slow drifts caused by natural evapotranspiration and crop absorption. Directly processing multidimensional observation sequences in a uniform manner can easily smooth out the impact characteristics of irrigation and fertilization or amplify stable noise. Using fixed thresholds or single agronomic indicators to judge water and fertilizer deficits does not standardize the deviation calculation for the optimal requirements and normal fluctuation range of taro at different growth stages, and cannot accurately reflect the stress level of each water-nutrient indicator. Conventional optimization algorithms do not consider the antagonistic effects between nitrogen, phosphorus, and potassium nutrients, and make decisions based solely on the deviation of a single nutrient indicator, which can easily generate water and fertilizer schemes with high nitrogen and low potassium, leading to nutrient synergy imbalance. Conventional optimization algorithms usually use fixed crossover and mutation probabilities, which cannot enhance the ability to explore new schemes when the water and fertilizer status in the root zone deteriorates rapidly. Furthermore, when constructing the fitness function, they do not simultaneously incorporate deviation improvement, nutrient synergy risk, and the cost of irrigation and fertilization inputs, making it difficult to balance agronomic effects, resource conservation, and nutrient balance. Summary of the Invention

[0004] To address the aforementioned issues, this invention proposes a taro water and fertilizer co-optimization method based on digital IoT data acquisition. Using a multidimensional original observation sequence of the taro root zone and an improved genetic algorithm, the optimal water and fertilizer decision vector is found to solve the water and fertilizer ratio problem during taro cultivation.

[0005] To achieve the above effects, the present invention adopts the following core concept: S1. Obtain multidimensional data of taro root zone, calculate the instantaneous change amplitude of adjacent points to mark pulse intervals, divide irrigation pulse intervals and non-pulse stable intervals, perform data processing on the observation indicators in the irrigation pulse intervals and non-pulse stable intervals to obtain irrigation pulse response values ​​and stable scale transformation values, and then generate event-driven scale transformation values ​​and agronomic state correction values. S2. Determine the current growth stage of taro, and call the optimal target values ​​of water and nutrients corresponding to the current growth stage and the normal relative fluctuation coefficient to calculate the standardized deviation of the agronomic state correction value, so as to obtain the water-nutrient coupling deviation characteristics. S3. Using soil available potassium concentration as a reference sequence for potassium absorption status in the root zone, and soil nitrate nitrogen concentration and soil available phosphorus concentration as comparison sequences, a nitrogen-phosphorus-potassium antagonism strength factor was constructed based on sliding grey relational analysis. S4. An improved genetic algorithm is used to find the optimal water and fertilizer decision vector. The improvements to the genetic algorithm include: a fitness function that integrates deviation, antagonism factor and input penalty, adaptive crossover mutation probability adjustment based on deviation deterioration rate, chaotic-reverse initialization and elite retention mechanism to update the population.

[0006] As a further technical solution, S1 specifically includes: Acquire multidimensional raw observation sequences of taro root zone, irrigation valve opening signals, and fertilizer pump operation signals; First-order difference calculations were performed on adjacent sampling points of the taro root zone observation index to obtain the instantaneous change amplitude of the index. Pulse interval markings were obtained based on irrigation valve opening signals, fertilizer pump operation signals, and instantaneous change amplitudes. Based on the pulse interval marking, the irrigation pulse interval and non-pulse stationary interval are divided. The event benchmark centering process is performed on the taro root area observation index in the irrigation pulse interval to obtain the irrigation pulse response value. The exponential moving average and minimum-maximum normalization process is performed on the taro root area observation index in the non-pulse stationary interval to obtain the stationary scale transformation value. Event-driven scale transformation values ​​and agronomic state correction values ​​are generated based on pulse interval markings.

[0007] As a further technical solution, S2 specifically includes: The number of days the taro grows is calculated based on the planting date and the current date, and the growth period markers of the taro are determined in conjunction with field phenological records; Based on the taro growth stage identifier, the water and fertilizer target parameter table is called to obtain the optimal target value and normal relative fluctuation coefficient of the water-nutrient index in the current growth stage; The agronomical state correction values ​​of the water-nutrient index are compared with the optimal target values ​​for the current growth period. The relative deviation is calculated first, and then divided by the normal relative fluctuation coefficient to obtain the standardized deviation. The standardized deviations corresponding to soil volumetric water content, soil nitrate nitrogen concentration, soil available phosphorus concentration, and soil available potassium concentration are arranged in a fixed order to form a water-nutrient coupling deviation vector.

[0008] As a further technical solution, S3 specifically includes: A sliding grey relational time window was selected, and soil nitrate nitrogen concentration sequence, soil available phosphorus concentration sequence, and soil available potassium concentration sequence were extracted within the sliding grey relational time window and then normalized. Using the normalized soil available potassium concentration sequence as a reference sequence, the point-by-point differences between the normalized soil nitrate nitrogen concentration sequence and the soil available potassium concentration sequence, as well as the point-by-point differences between the normalized soil available phosphorus concentration sequence and the soil available potassium concentration sequence, were calculated to obtain the nitrogen-potassium correlation coefficient sequence and the phosphorus-potassium correlation coefficient sequence, respectively. The nitrogen-potassium correlation coefficient sequence and the phosphorus-potassium correlation coefficient sequence are averaged to obtain the nitrogen-potassium grey correlation degree and the phosphorus-potassium grey correlation degree, respectively. The nitrogen-phosphorus-potassium antagonism strength factor is calculated based on the nitrogen-potassium grey relational degree and the phosphorus-potassium grey relational degree.

[0009] As a further technical solution, the fitness function in S4 specifically includes: Candidate water and fertilizer decision vectors are constructed based on single irrigation duration, single nitrogen fertilizer application rate, single phosphate fertilizer application rate, and single potash fertilizer application rate. Based on historical irrigation test data, the unit water and fertilizer input response coefficient matrix is ​​pre-calibrated, and a unit water and fertilizer input response model is constructed based on the unit water and fertilizer input response coefficient matrix. Based on the candidate water and fertilizer decision vectors, the expected deviation vector after application is obtained using a response model; Adaptive importance weights are determined based on the current growth stage of taro and recent deviations. Estimate the expected nitrogen, phosphorus, and potassium antagonistic strength factor based on the current nitrogen, phosphorus, and potassium antagonistic strength factor and candidate water and fertilizer decision vectors; A fitness function is constructed by integrating expected deviation, nitrogen-potassium ratio penalty, irrigation input penalty, and fertilizer input penalty.

[0010] As a further technical solution, the S4 crossover mutation probability adjustment specifically includes: The rate of deterioration of water deviation is the difference between the average expected water deviation of two adjacent generations of population. The average expected moisture deviation is obtained by averaging the expected soil volume moisture deviations corresponding to all candidate water and fertilizer decision vectors in the statistical population. The ratio of the rate of deterioration of moisture deviation to the threshold of the rate of deterioration of moisture deviation is calculated and truncated to a non-negative value to obtain the deterioration intensity adjustment amount; The crossover and mutation probabilities are updated based on the evolutionary process and the degree of deterioration.

[0011] As a further technical solution, S4 chaos-reverse initialization specifically includes: Set the boundaries of water and fertilizer decision variables, and generate initial chaotic variables in the range of 0 to 1 for each candidate solution and each water and fertilizer decision variable; Chaotic mapping iterative update is adopted to map the chaotic variables to their actual values ​​according to the upper and lower bounds of the corresponding water and fertilizer decision variables, thereby obtaining chaotic candidate solutions; For each chaotic candidate solution, construct an inverse candidate solution. Then merge the chaotic candidate solutions and the inverse candidate solutions to obtain the initial candidate solution set.

[0012] As a further technical solution, the S4 elite retention mechanism includes: Simulated binary crossover is performed based on the crossover probability, and polynomial mutation is performed based on the mutation probability. An elite preservation mechanism is used to update the population, merging the parent population and the offspring population after crossover mutation to form a mixed population; In the mixed population, the fitness values ​​of all individuals are calculated and sorted from smallest to largest. The half of the individuals with the smaller fitness values ​​are taken as the next generation population. The algorithm determines whether to proceed to the next iteration or output the final result based on the stopping condition.

[0013] As a further technical solution, the algorithm stopping conditions include: The algorithm stops when the improvement in the optimal fitness value is less than the set value for 10 consecutive generations or when the number of iterations reaches the maximum value. The vector corresponding to the individual with the smallest fitness value in the current population is output as the optimal water and fertilizer decision vector. Otherwise, the next round of the loop is performed.

[0014] Beneficial effects of this invention: First, this invention identifies irrigation pulse intervals by jointly using irrigation event-driven and observed index mutation intensity analysis. It then performs event benchmark centering on pulse intervals and exponential moving average and minimum-maximum normalization on non-pulse stationary intervals, generating event-driven scaling values ​​and agronomic state correction values. This decouples the preservation of irrigation impact amplitude with the suppression of stationary noise. Second, based on the optimal target values ​​of water and nutrients dynamically allocated during the current growth stage of taro and their normal relative fluctuation coefficients, the multidimensional agronomic state correction values ​​are transformed into standardized deviations and combined into a water-nutrient coupled deviation vector, eliminating the influence of different index dimensions and natural fluctuations. The impact of amplitude differences was investigated. Then, using the soil available potassium concentration as a reference sequence, the gray correlation between nitrogen and potassium and between phosphorus and potassium were calculated within a sliding time window. The decrease in correlation was transformed into a nitrogen-phosphorus-potassium antagonism strength factor, quantifying the risk of synergistic imbalance in nitrogen and phosphorus absorption of potassium. Finally, an adaptive fitness function was constructed that integrates water-nutrient deviation cost, nitrogen-potassium ratio penalty, irrigation input penalty, and fertilization input penalty. The crossover probability and mutation probability of the genetic algorithm were dynamically adjusted in conjunction with the deviation deterioration rate. At the same time, a chaotic-reverse elite initialization and elite retention mechanism was adopted to realize the global optimization search of water and fertilizer schemes. Attached Figure Description

[0015] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0016] Figure 1 This is a schematic diagram of the process of the present invention; Figure 2 This is a line graph illustrating the event-driven pulse interval identification of the present invention. Figure 3 This is a schematic diagram illustrating the extraction of water-nutrient coupling deviation features based on dynamic matching of taro growth period according to the present invention; Figure 4 This is a schematic diagram illustrating the dynamic quantification of the nitrogen-phosphorus-potassium antagonism strength factor of taro based on sliding gray correlation in this invention. Figure 5 This is a line graph illustrating the dynamic quantification of the sliding grey relational antagonistic factor in this invention. Figure 6 This is a schematic diagram of the adaptive crossover mutation probability adjustment based on the deviation deterioration rate of the present invention; Figure 7 This is a schematic diagram of the chaotic-reverse initialization, elite retention mechanism and polynomial mutation offspring generation of the present invention; Figure 8 This is a line graph showing the convergence process of the genetic algorithm in this invention. Figure 9 This is a bar chart showing the fitness cost decomposition of the candidate water and fertilizer schemes for this invention. Detailed Implementation

[0017] The technical solutions of the embodiments 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, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] 1. Adaptive scaling transformation and decoupling of multidimensional data from taro root region The synergistic management of water and fertilizer in taro relies on multi-source Internet of Things data, such as soil volumetric water content, soil electrical conductivity, soil nitrate nitrogen concentration, soil available phosphorus concentration, soil available potassium concentration, relative humidity, and photosynthetically active radiation. The opening of irrigation valves and the operation of fertilizer pumps will cause short-term step changes in water and salinity indicators in the taro root zone. Natural evapotranspiration, crop absorption, and environmental changes will cause these indicators to drift slowly.

[0019] This step uses irrigation valve opening signals and fertilizer pump operation signals as event triggers, and combines the first-order difference fluctuation intensity of the multidimensional original observation sequence of taro root area to identify irrigation pulse intervals. Event benchmark centering is performed on the irrigation pulse intervals, and exponential moving average and stationary window normalization are performed on non-pulse stationary intervals. This process preserves the irrigation impact amplitude while suppressing high-frequency noise in stationary intervals. The specific steps are as follows: (1) The multidimensional original observation sequence of the taro root region obtained through IoT acquisition devices can be represented as: in, Indicates the first The multidimensional original observation vector of the taro root region at any given time is used to characterize the water, salinity, nutrient and environmental status of the taro root region; Indicates the first The observation indicators of the taro root area were in the first The original observation value at that time; This represents the index of observation indicators in the taro root region, with values ​​ranging from 1 to... ; This represents the total number of observation indicators in the taro root zone, which can be taken as 7, corresponding to soil volumetric water content (in %), soil electrical conductivity (in dS / m), soil nitrate nitrogen concentration (in mg / kg), soil available phosphorus concentration (in mg / kg), soil available potassium concentration (in mg / kg), air relative humidity (in %), and photosynthetically active radiation (in μmol / (m²·s)). This represents the sampling time index, used to locate the time position when the IoT terminal records data at a fixed sampling interval, which can be 10 minutes. This represents the transpose operation of a vector.

[0020] Then, the irrigation valve opening signal and the fertilizer pump operation signal are acquired synchronously. The irrigation valve opening signal is used to indicate whether the irrigation system is in the water supply state, and the fertilizer pump operation signal is used to indicate whether the fertilizer system is in the fertilizer injection state. When the irrigation valve opening signal or the fertilizer pump operation signal is in the triggered state, the system enters the irrigation event candidate judgment stage.

[0021] (2) For the first First-order difference calculations were performed on adjacent sampling points of the taro root region observation index to obtain the index at the [missing information - likely a specific location or point]. The instantaneous change amplitude at time t is used to reflect the degree of abrupt change of the current sampling point relative to the previous sampling point; then, in the most recent The moving standard deviation of the first-order difference series within each sampling point. , Indicates the first The observation indicators of the taro root area were in the first The sliding standard deviation of fluctuation at any given time is used to characterize the intensity of natural fluctuations within the current local time range. This represents the length of the sliding statistical window, used to determine the range of local fluctuation estimation. It can be set to 12, which corresponds to an observation range of 120 minutes when the sampling interval is 10 minutes.

[0022] If the irrigation valve opening signal or the fertilizer pump running signal is in the triggered state, and the first The instantaneous change amplitude of the observed indicators in the taro root area is greater than If the above conditions are met, the sampling point is marked as an irrigation pulse sampling point; otherwise, it is marked as a non-pulse stationary sampling point. This yields the pulse interval marking. , Indicates the first The observation indicators of the taro root area were in the first The pulse interval marker for a given time is set to 1, indicating that the time belongs to the irrigation pulse interval, and 0, indicating that the time belongs to the non-pulse steady interval.

[0023] As an example, if the sliding statistical window length The current raw observation value is 22.1%, and the previous raw observation value was 21.5%. Therefore, the instantaneous change amplitude at this moment is... , i.e., 0.60; the first-order difference sequence of the most recent 12 sampling points of this index is calculated as follows: The calculated standard deviation of the moving average fluctuation of the sequence is: At this time, the irrigation valve opening signal is in the triggered state, and the instantaneous change amplitude is greater than 0.60. Therefore, this sampling point is marked as the irrigation pulse sampling point.

[0024] The irrigation pulse interval is not determined solely by the irrigation valve opening signal or the fertilizer pump running signal, but is jointly determined by the equipment triggering status and the intensity of sudden changes in the observed indicators. This avoids situations where the valve is open but the root zone has not yet responded, or the fertilizer pump is running but the sensor reading has not changed effectively, which may be misjudged as valid irrigation pulses.

[0025] (3) Within the irrigation pulse interval, i.e. when At that time, for the first The observation indicators of each taro root zone were processed using event-based benchmarking to obtain the irrigation impulse response values.

[0026] Specifically, a stable sampling window is selected before the irrigation event is triggered. The length of the stable sampling window can be 6 to 12 sampling points. After removing obvious outliers, the average of the original observations is calculated to obtain the baseline mean of the event. , Indicates the first The observation indicators of the taro root area were in the first The baseline mean of the event at any given time is used to represent the basic level of the indicator before the irrigation event occurs. Further, the irrigation impulse response value is obtained by subtracting the baseline mean from the current raw observation value. This value is used to retain the true impact amplitude caused by irrigation or fertilization.

[0027] As an example, if the soil volumetric moisture content is stable at 21.8% before the irrigation valve is opened, and the soil volumetric moisture content rises to 26.5% after the irrigation response, the irrigation pulse response value obtained by the event baseline centering process is 4.7%, which directly reflects the increase in soil volumetric moisture content in the taro root zone by a single irrigation.

[0028] (4) In the non-pulse stationary region, i.e. when At that time, for the first The observation indicators of each taro root region were processed by exponential moving average and min-maximum normalization to obtain stationary scale transformation values.

[0029] Specifically, the exponential moving average process updates the data by weighting the current original observations and the smoothed values ​​from the previous time step. The update method is expressed as follows: in, Indicates the first The observation indicators of the taro root area were in the first The exponential moving average of time is used to suppress sensing noise in non-pulse stationary regions; Indicates the first The observation indicators of the taro root area were in the first The exponential moving average at time t; the weights of the current raw observations are determined by the exponential moving average coefficient. control, This indicates the strength of the influence of the current original observations on the smoothing result. The larger the value, the faster the smoothing result responds to the latest data. A value of 0.3 is acceptable.

[0030] Then, in the non-pulse stationary region, the nearest Within each sampling point, the minimum value of the stationary window and the maximum value of the impulse window are statistically analyzed. Then, min-max normalization is performed on the smoothed values ​​of the stationary state to obtain the stationary scaling transformation values. The value represents the length of the stationary normalization window, used to limit the range of normalization statistics and avoid the influence of early historical data on the current root region state judgment. It can be 72, which corresponds to a 12-hour observation range when the sampling interval is 10 minutes. The minimum value of the stationary window is used to determine the lower bound of the stationary normalization, and the maximum value of the stationary window is used to determine the upper bound of the stationary normalization.

[0031] As an example, if we are currently in a non-pulse stationary range, for the soil volumetric water content index, at the current moment... raw observations It is 21.5%, the smoothed value of the previous time step. It is 21.2%, the exponential moving average coefficient. Taking 0.3, the exponential moving average at the current time is... If in the most recent Within each sampling point, the minimum smoothed value of this index is 20.8%, and the maximum value is 23.0%. After min-maximum normalization, the resulting stationary scaling transformation value is... .

[0032] (5) According to the pulse interval marking Generate event-driven scale transformation values , Indicates the first The observation indicators of the taro root area were in the first The event-driven scaling transformation value at time step.

[0033] Specifically, when hour, The event-based, benchmark-centered irrigation impulse response value is used to preserve the irrigation impact intensity; when hour, The stationary scaling transformation value is used to eliminate the differences in the dimensions of different indicators and reduce stationary noise.

[0034] (6) Generate agronomic state correction values ​​based on pulse interval markers. , Indicates the first The observation indicators of the taro root area were in the first Corrected observations used for comparing agronomic objectives at specific times.

[0035] Specifically, within the irrigation pulse interval, Take the original observations after outlier removal; in the non-pulse stationary interval, Take the exponential moving average.

[0036] Analyze event-driven pulse interval identification line graphs, such as Figure 2 As shown, this figure illustrates the change in soil volumetric water content over time and the identification of irrigation pulse events. The horizontal axis represents time (in minutes), and the vertical axis represents soil volumetric water content (in %). The original observation value is represented by a blue broken line, the irrigation pulse interval is marked by an orange background, the red scatter dots represent irrigation pulse sampling points, and the green scatter dots represent non-pulse stationary sampling points. The experiment shows that when the irrigation valve is opened or the fertilizer pump is running, the soil volumetric water content shows a significant instantaneous step increase, while under natural conditions, there are only small, stable fluctuations.

[0037] The event-driven scaling transformation value in this step is used to preserve the irrigation response characteristics, and the agronomic state correction value is used to compare with the optimal target value for the current growth period, avoiding the inconsistency of dimensions caused by directly comparing the centered pulse amplitude with the agronomic target value.

[0038] 2. Feature extraction of water-nutrient coupling deviation based on dynamic matching of taro growth period Because taro goes through germination, seedling, stem and leaf growth, corm enlargement, and maturity stages from planting to harvest, the requirements for soil volumetric water content, soil nitrate nitrogen concentration, soil available phosphorus concentration, and soil available potassium concentration vary significantly at each growth stage. This step determines the current growth stage of the taro based on the current date, planting date, and field phenological records. It then uses the optimal target values ​​for water and nutrients corresponding to the current growth stage, along with normal fluctuation coefficients, to calculate the standardized deviation of agronomic state correction values. This yields directly comparable water-nutrient coupling deviation characteristics, thus unifying the multi-dimensional, multi-dimensional root zone water and fertilizer status into dimensionless indicators, providing standardized input for subsequent optimization decisions. The specific process is as follows: Figure 3 As shown, the detailed steps are as follows: (1) Calculate the number of days the taro grows based on the planting date and the current date, and determine the taro growth period markers by combining the field phenological records. , Indicates the first The corresponding taro growth stage identifier is used to call the water and fertilizer target parameters for different growth stages, including the germination stage, seedling stage, stem and leaf growth stage, corm enlargement stage, and maturity stage.

[0039] In practice, if the growth period is determined solely by the number of days of growth, the range of days for each growth period can be pre-set according to the local cultivation system. If there are conditions such as low temperature, rain, or delayed emergence in the field, the growth period label can be corrected by the phenological conditions obtained through manual recording or image recognition, so that the taro growth period label is more consistent with the actual growth conditions in the field.

[0040] (2) Based on the taro growth period identification Calling the water and fertilizer target parameter table, we obtain the first... The optimal target values ​​for each water-nutrient index during the current growth stage. and normal relative volatility coefficient .

[0041] Among them, the moisture-nutrient index The values ​​in this step and subsequent steps range from 1 to 4. The first water-nutrient index is the soil volumetric water content, the second water-nutrient index is the soil nitrate nitrogen concentration, the third water-nutrient index is the soil available phosphorus concentration, and the fourth water-nutrient index is the soil available potassium concentration. Indicates the first The optimal target values ​​for each water-nutrient index at the current growth stage are obtained from the taro cultivation agronomic parameter table, historical high-yield field data, or local experimental data, and are used to provide the optimization target center for the current stage. Indicates the first The normal relative fluctuation coefficients of water-nutrient indicators during the current growth period are obtained from the relative fluctuation statistics of normal management samples in the same period in history. They are used to eliminate the deviation scale differences caused by the different natural fluctuation amplitudes of different water-nutrient indicators.

[0042] The normal relative fluctuation coefficient is a statistical measure of the allowable fluctuation range of an indicator under normal water and fertilizer management conditions during the current growth period. The smaller the normal relative fluctuation coefficient, the more sensitive the indicator is to deviation during the current growth period, and the higher the standardized deviation will be for the same relative deviation.

[0043] (3) The first Agronomic state correction values ​​for each moisture-nutrient index By comparing the value with the optimal target value for the current reproductive period, the relative deviation is first calculated, and then divided by the normal relative fluctuation coefficient to obtain the standardized deviation. The calculation method is expressed as follows: in, Indicates the first The water-nutrient index in the first The standardized deviation at any given time is used to measure the degree of deviation of this indicator from the agronomically suitable range for the current growth period. The larger the value, the higher the risk of water and fertilizer stress. Indicates the first The water-nutrient index in the first The agronomic state correction value at any given time is used to provide noise-suppressed input for deviation calculation; This indicates the absolute value operation, used to ensure that the standardized deviation is non-negative.

[0044] (4) The standardized deviations corresponding to soil volumetric water content, soil nitrate nitrogen concentration, soil available phosphorus concentration, and soil available potassium concentration are arranged in a fixed order to form a water-nutrient coupling deviation vector. , is represented as: in, Indicates the first The moisture-nutrient coupling deviation vector at time step, with dimension . It is used to simultaneously characterize the stress levels of water, nitrogen, phosphorus and potassium in the root zone of taro; This represents the transpose operation of a vector.

[0045] As an example, when taro is in the corm enlargement stage, if the optimal target value for soil volumetric water content is set to 27.5% and the normal relative fluctuation coefficient is set to 0.08, and the optimal target value for soil available potassium concentration is set to 180 mg / kg and the normal relative fluctuation coefficient is set to 0.10, when the agronomically corrected value of soil volumetric water content is lower than the optimal target value, the standardized deviation of soil volumetric water content increases, and the subsequent genetic algorithm will prioritize increasing the duration of a single irrigation. When the soil available potassium concentration is lower than the optimal target value and the soil nitrate nitrogen concentration is high, the subsequent fitness function will simultaneously increase the potassium supplementation requirement and the nitrogen-potassium antagonism penalty, thereby avoiding the generation of a high-nitrogen, low-potassium water and fertilizer scheme.

[0046] 3. Dynamic Quantification of Nitrogen-Phosphorus-Potassium Antagonism Intensity Factor in Taro Based on Sliding Grey Relation Since nitrogen, phosphorus, and potassium nutrients in the taro root zone are not independent, excessive nitrogen can inhibit potassium absorption, and abnormal phosphorus levels can also alter potassium availability. This step uses soil available potassium concentration as a reference sequence for root zone potassium absorption status, and soil nitrate nitrogen and available phosphorus concentrations as comparison sequences. Within a sliding time window, the gray correlation between nitrogen and potassium and between phosphorus and potassium is calculated. The decrease in correlation is then converted into a nitrogen-phosphorus-potassium antagonism strength factor, quantifying the potential inhibitory risk to potassium absorption caused by excessive or unbalanced nitrogen and phosphorus nutrients. This addresses the shortcomings of relying solely on a single indicator threshold. The specific process is as follows: Figure 4 As shown, a detailed description follows: (1) in the Select the associated time window by moving forward in time.

[0047] Specifically, the length of the sliding gray association time window is denoted as... , This represents the number of sampling points within the sliding gray correlation time window, used to limit the dynamic analysis range of the nitrogen, phosphorus, and potassium synergistic relationship. It can be set to 144, which corresponds to a 24-hour observation range when the sampling interval is 10 minutes.

[0048] Then, soil nitrate nitrogen concentration sequence, soil available phosphorus concentration sequence, and soil available potassium concentration sequence were extracted within the sliding grey correlation time window, and min-max normalization was performed on each of them to convert each type of nutrient concentration sequence within the sliding grey correlation time window into a range of 0 to 1, thereby making different nutrient concentration sequences within a comparable numerical range.

[0049] (2) Using the normalized soil available potassium concentration sequence as the reference sequence, calculate the point-by-point difference between the normalized soil nitrate nitrogen concentration sequence and the soil available potassium concentration sequence, as well as the point-by-point difference between the normalized soil available phosphorus concentration sequence and the soil available potassium concentration sequence.

[0050] Specifically, the smaller the difference at each point, the more consistent the corresponding nutrient change trend is with the soil available potassium concentration change trend; the larger the difference at each point, the less consistent the corresponding nutrient change trend is with the soil available potassium concentration change trend.

[0051] First, within the sliding grey relational analysis time window, the minimum and maximum values ​​of all point-by-point differences are statistically analyzed. For each sampling point within the window, the point-by-point difference value is obtained. Then, the minimum difference value is multiplied by the grey relational analysis resolution coefficient and the maximum difference value to obtain the numerator, and the point-by-point difference value is multiplied by the grey relational analysis resolution coefficient and the maximum difference value to obtain the denominator. Dividing these two values ​​yields the grey relational coefficient for that sampling point. Finally, the grey relational coefficients are combined in order consistent with the soil available potassium concentration sequence to obtain the nitrogen-potassium correlation coefficient sequence and the phosphorus-potassium correlation coefficient sequence, respectively. The correlation coefficients for all sampling points are between 0 and 1. The smaller the point-by-point difference, the closer the correlation coefficient is to 1, indicating that the trend of the comparison sequence at that point is more consistent with the reference sequence. The grey relational analysis resolution coefficient is used to adjust the difference resolution capability and can be set to 0.5.

[0052] (3) Calculate the average of the nitrogen-potassium correlation coefficient series to obtain the nitrogen-potassium grey correlation degree. , Indicates the first The nitrogen-potassium grey correlation degree at time point is used to measure the consistency between changes in soil nitrate nitrogen concentration and changes in soil available potassium concentration within a sliding grey correlation time window. The smaller the value, the less consistent the relationship between nitrogen and potassium changes.

[0053] Then, the phosphorus-potassium correlation coefficient sequence is averaged to obtain the phosphorus-potassium grey correlation degree. , Indicates the first The phosphorus-potassium grey correlation degree at time point is used to measure the consistency between changes in soil available phosphorus concentration and changes in soil available potassium concentration within a sliding grey correlation time window. The smaller the value, the less consistent the relationship between phosphorus and potassium changes.

[0054] (4) Calculate the nitrogen-phosphorus-potassium antagonism strength factor based on the nitrogen-potassium grey relational degree and the phosphorus-potassium grey relational degree. .

[0055] Specifically, first, the reciprocals of the gray correlation coefficients for nitrogen-potassium and phosphorus-potassium are calculated separately, so that a decrease in correlation coefficient can be converted into an increase in antagonistic risk. Then, the mean of the two reciprocal results is taken, and the antagonistic intensity benchmark shift coefficient is subtracted before natural exponential mapping. Finally, the nitrogen-phosphorus-potassium antagonistic intensity factor is obtained through amplitude limiting processing. The calculation method is expressed as follows: in, Indicates the first The nitrogen-phosphorus-potassium antagonism strength factor at a given time is used to quantify the potential inhibitory risk of nitrogen and phosphorus on potassium availability; the higher the value, the higher the risk of nutrient synergistic imbalance. This represents the baseline shift coefficient for antagonistic strength, used to bring the nitrogen-phosphorus-potassium antagonistic strength factor close to 1 when nutrient correlation is normal; it can be set to 1.0. This represents the lower limit of the nitrogen-phosphorus-potassium antagonism strength factor, used to avoid the penalty term being too small, which would cause the nutrient synergistic constraint to fail. It can be taken as 0.8. This represents the upper limit of the nitrogen-phosphorus-potassium antagonism strength factor, used to avoid excessive penalty in the fitness function due to abnormal sensor data; a value of 5.0 is acceptable. This represents the limiting operator, used to restrict the calculation result to a certain range. to between; This represents the natural exponential function, used to nonlinearly amplify the antagonistic risk caused by a decrease in correlation.

[0056] By integrating two independent grey relational degrees into a single antagonistic risk factor, the antagonistic effect of nitrogen and phosphorus nutrients on potassium absorption is quantified into a scalar factor, providing real-time feedback for the fitness function to reflect the risk of soil nutrient synergy. When the nitrogen-potassium grey relational degree or the phosphorus-potassium grey relational degree is abnormally close to 0, the lower limit of the relational degree can be set to 0.05 before participating in the reciprocal calculation to avoid distortion of the antagonistic intensity factor caused by extreme outlier data.

[0057] As an example, if the sampling interval is 10 minutes and the sliding gray correlation time window length is 144, corresponding to the data of the past 24 hours; in the... At any given time, the soil nitrate nitrogen, available phosphorus, and available potassium concentration sequences within the window are extracted and normalized. Using available potassium as the reference sequence, the minimum difference for all points within the window is calculated to be 0.01, and the maximum is 0.45, with a resolution coefficient of 0.5. For a given sampling point, if the normalized difference between nitrate nitrogen and available potassium is 0.1, then the nitrogen-potassium grey relational coefficient for that point is... The nitrogen-potassium grey relational degree is obtained by averaging the correlation coefficients of the 144 points within the window. Phosphorus-potassium grey relational degree If the baseline shift factor for antagonistic strength is taken as 1.0, the lower limit of the nitrogen-phosphorus-potassium antagonistic strength factor is taken as 0.8, and the upper limit of the nitrogen-phosphorus-potassium antagonistic strength factor is taken as 5.0, then the nitrogen-phosphorus-potassium antagonistic strength factor is: This indicates that there is currently a certain degree of risk of nutrient imbalance.

[0058] Analyze the dynamic quantitative line graph of the sliding grey relational antagonistic factor, such as Figure 5As shown, the dynamic changes of nitrogen-potassium grey relational degree, phosphorus-potassium grey relational degree, and nitrogen-phosphorus-potassium antagonistic strength factor within a sliding time window are illustrated. The horizontal axis represents time (in minutes), the left vertical axis represents grey relational degree (dimensionless), and the right vertical axis represents nitrogen-phosphorus-potassium antagonistic strength factor (dimensionless). In the figure, the green broken line represents the nitrogen-potassium grey relational degree, the blue broken line represents the phosphorus-potassium grey relational degree, and the red broken line represents the nitrogen-phosphorus-potassium antagonistic strength factor. The experiment shows that when the grey relational degree decreases, the antagonistic strength factor exhibits a non-linear increase, proving that the sliding grey relational method can effectively quantify the risk of synergistic imbalance in nitrogen and phosphorus absorption of potassium.

[0059] 4. Construction of an adaptive fitness function that integrates deviation, antagonistic factor, and input penalty. Because taro irrigation and fertilization decisions need to consider root zone water and nutrient status, the risk of nitrogen, phosphorus, and potassium synergistic absorption, and the cost of water and fertilizer input, this step uses the water-nutrient coupling deviation vector and the nitrogen, phosphorus, and potassium antagonism intensity factor as core inputs. Candidate irrigation and fertilization decisions are substituted into a pre-calibrated unit water and fertilizer input response model to predict the deviation change after application. A fitness function is constructed by integrating nitrogen and potassium ratio penalties, irrigation input penalties, and fertilizer input penalties, thereby quantifying the comprehensive benefits of any candidate irrigation and fertilization scheme. This achieves a balance between improving root zone stress, avoiding nutrient antagonism, and saving input costs, avoiding resource waste or nutrient imbalance caused by unilaterally pursuing optimal agronomic states. This provides a clear search direction for the genetic algorithm. The specific steps are as follows: (1) Construct candidate water and fertilizer decision vectors , is represented as: in, Represents the candidate water and fertilizer decision vector, with dimension . , used to characterize candidate water and fertilizer schemes in genetic algorithms; This indicates the duration of a single irrigation session, in minutes, and is used to control the valve opening time of the irrigation system. The value ranges from 0 to 90. This indicates the amount of nitrogen fertilizer applied in a single application, expressed in kg / ha, and is used to control the amount of nitrogen fertilizer injected. This indicates the amount of phosphate fertilizer applied in a single application, expressed in kg / ha, and is used to control the amount of phosphate fertilizer injected. This indicates the amount of potassium fertilizer applied in a single application, expressed in kg / ha, and is used to control the amount of potassium fertilizer injected. This represents the transpose operation of a vector.

[0060] (2) Pre-calibrate the unit water and fertilizer input response coefficient matrix based on historical irrigation test data. , This represents the response coefficient matrix per unit water and fertilizer input, with dimension 1. It is used to describe the ability of a single irrigation duration, a single nitrogen fertilizer application rate, a single phosphate fertilizer application rate, and a single potassium fertilizer application rate to improve the deviation of soil volumetric water content, soil nitrate nitrogen concentration, soil available phosphorus concentration, and soil available potassium concentration.

[0061] Specifically, different water and fertilizer treatment combinations with different irrigation durations, nitrogen fertilizer application rates, phosphorus fertilizer application rates, and potassium fertilizer application rates can be set up in the taro experimental field. After each irrigation is completed and the root zone response stabilizes, the change in the water-nutrient coupling deviation vector is recorded. Then, with the candidate water and fertilizer input as the independent variable and the reduction in deviation as the dependent variable, the unit water and fertilizer input response coefficient matrix is ​​fitted using the non-negative constraint least squares method.

[0062] Among them, the nonnegative constraint least squares method is a parameter estimation method. It uses an optimization algorithm to find a set of unit water and fertilizer input response coefficient matrices that minimizes the sum of squared errors between the predicted values ​​(i.e., the product of the unit water and fertilizer input response coefficient matrix to be estimated and the candidate water and fertilizer decision vector) and the actual values ​​across all dimensions. The nonnegative constraint means that all coefficients in the matrix are greater than or equal to 0, ensuring that the improvement direction of irrigation and fertilization on the corresponding deficiency type deviation conforms to agronomic logic; that is, increasing the amount of irrigation and fertilization will only reduce or maintain the deviation.

[0063] (3) The candidate water and fertilizer decision vectors Substitute the values ​​into the unit water and fertilizer input response model to predict the expected deviation vector after application. .

[0064] Specifically, based on the unit water and fertilizer input response coefficient matrix Construct a unit water and fertilizer input response model by using the current deviation vector Subtract the candidate water and fertilizer decision vector Improvement in deviation The expected deviation vector is obtained. Its input is the candidate water and fertilizer decision vector. The output is the expected deviation vector. .

[0065] Among them, the unit water and fertilizer input response model is used to predict the implementation of a candidate water and fertilizer scheme. Subsequently, the expected decrease in the deviation of each water-nutrient index is calculated, specifically by determining the expected deviation vector. ,Right now: In the formula, This represents the expected deviation vector corresponding to the candidate water and fertilizer decision vector, with dimension 1. It is used to simultaneously characterize the improvement effect of candidate fertilization and irrigation schemes on deviations in water, nitrogen, phosphorus and potassium. Represents the first candidate water and fertilizer decision vector. The expected deviation of a moisture-nutrient index is used to predict the residual stress level of that index at the end of application. The dimension is A zero vector is used to limit the expected deviation to be no less than 0; This represents the maximum value function, used to ensure that the deviation of the prediction does not become negative.

[0066] (4) Determine the adaptive importance weight based on the current growth stage of taro and recent deviation changes. .

[0067] First, calculate the... Several indicators in recent Deviation trend of individual sampling points This can be simplified to the most recent The slope of the standardized deviation linear regression (the linear regression slope refers to the rate of increase or decrease of the dependent variable with respect to each sampling interval after fitting an optimal straight line using the least squares method to a set of data points arranged in time or order); then, the original weights are calculated. The calculation method is expressed as follows: Finally, normalization is performed so that the sum of all weights is 1, which can be expressed as: in, Indicates the first The water-nutrient index in the first The adaptive importance weights at different times are used to adjust the influence of each indicator on the fitness value under different reproductive stages and different stress trends. This indicates the taro's growth period labeling. The first call The basic weights of each indicator (the basic weights can be obtained by consulting or calibrating the water and fertilizer management priority parameter table for different growth stages of taro). This indicates the length of the weight update window, used to statistically analyze recent deviation trends, and can be taken as 6 to 12 sampling points.

[0068] Adaptive importance weights enable the fitness function to prioritize correcting deteriorating root zone states. For example, when the soil available potassium concentration is low and the deviation continues to increase during the bulb enlargement period, the system increases the weight of soil available potassium concentration, making the genetic algorithm more inclined to search for candidate water and fertilizer schemes with stronger potassium replenishment capabilities and a reasonable nitrogen-potassium ratio.

[0069] (5) Based on the current nitrogen-phosphorus-potassium antagonism strength factor and candidate water and fertilizer decision vectors Estimated expected nitrogen-phosphorus-potassium antagonism strength factor , The expected nitrogen-phosphorus-potassium antagonism strength factor corresponding to the candidate water and fertilizer decision vector is used to evaluate the risk of nutrient synergy imbalance that may be caused by the candidate water and fertilizer scheme.

[0070] Specifically, the nitrogen-potassium application ratio and phosphorus-potassium application ratio in the candidate water and fertilizer decision vectors can be calculated first. Then, the impact of the candidate water and fertilizer schemes on nutrient synergy can be estimated by querying the historical antagonistic mapping table or by using a piecewise linear mapping method, thereby obtaining the candidate water and fertilizer decision vectors. Corresponding expected nitrogen-phosphorus-potassium antagonistic strength factor For example, when Compared to When significantly elevated, the nitrogen-phosphorus-potassium antagonism strength factor is expected to increase from its current level; when When supplementation is adequate and the nitrogen-potassium application ratio is within the agronomic recommendation range, the nitrogen-phosphorus-potassium antagonism strength factor is not expected to increase further.

[0071] (6) Integrate expected deviation, nitrogen-potassium ratio penalty, irrigation input penalty and fertilizer input penalty to construct a fitness function. The calculation is expressed as: in, The fitness function represents the candidate water and fertilizer decision vector, which is used to evaluate the overall merits of the candidate water and fertilizer schemes. The genetic algorithm aims to minimize this value. This represents the antagonistic penalty coefficient, used to amplify the risks caused by an unreasonable nitrogen-potassium ratio; a value of 4.0 is acceptable. This represents the minimum reference amount of potassium fertilizer, used to avoid distortion in the calculation of the nitrogen-potassium application ratio when the candidate potassium fertilizer application amount is 0. It can be taken as 1 kg / ha. This represents the maximum recommended nitrogen-potassium ratio for agronomical applications, used to constrain the upper limit of nitrogen fertilizer input relative to potassium fertilizer input, and can be taken as 1.2; This represents the irrigation input penalty coefficient, used to suppress unnecessary prolonged irrigation; a value of 0.3 is acceptable. This indicates the upper limit of a single irrigation duration, which is determined by the capacity of the irrigation equipment, the soil infiltration capacity, and the field drainage conditions; 90 minutes is a suitable value. This represents the fertilizer input penalty coefficient, used to suppress unnecessary high fertilizer input, and can be set to 0.5; This represents the fertilizer type index, with a set of values. These correspond to nitrogen fertilizer, phosphorus fertilizer, and potassium fertilizer, respectively. Indicates the first The single application rate of fertilizers is used to uniformly represent the candidate input amounts of nitrogen, phosphorus, and potassium fertilizers. Indicates the first The upper limit of the single application amount of fertilizer is determined by the capacity of the fertilizer applicator, the injection capacity of the pipeline and the local fertilization regulations. It can be taken as 45 kg / ha for nitrogen fertilizer, 30 kg / ha for phosphorus fertilizer and 60 kg / ha for potassium fertilizer. This represents the maximum value function, used to trigger the antagonistic penalty only when the nitrogen-potassium application ratio exceeds a threshold.

[0072] It needs to be emphasized that, The item is the water-nutrient deviation cost item, which is used to measure the degree to which each nutrient index deviates from the optimal state after the implementation of the candidate water and fertilizer scheme; The term is a nitrogen-potassium antagonism penalty term, which applies when the nitrogen-potassium ratio of the candidate solution is... Exceeding the maximum nitrogen-potassium application ratio When the time comes, the punishment will be initiated; The item is the irrigation input cost item, which imposes a penalty on excessively long irrigation time in order to conserve water resources and prevent waterlogging. This item represents the cost of fertilizer input, which penalizes excessive amounts of fertilizer in order to save on fertilizer costs and reduce environmental risks.

[0073] 5. Adaptive crossover mutation probability adjustment based on deviation deterioration rate In scenarios such as high temperature and drought, sudden rainfall, or excessive fertilization, the water-nutrient deviation can deteriorate rapidly within a short period. This step uses the continuously output water-nutrient coupling deviation vector to calculate the deviation deterioration rate and dynamically adjusts the crossover and mutation probabilities of the genetic algorithm according to the degree of deterioration. This enhances the ability to explore new candidate water and fertilizer schemes when the root zone state deteriorates rapidly, and enhances the ability to perform fine-grained local search when the root zone state is stable. The specific process is as follows: Figure 6 As shown, a detailed description follows: (1) in the In the genetic algorithm, the expected deviation of soil volumetric water content corresponding to all candidate water and fertilizer decision vectors in the current generation is statistically analyzed, and the average expected water content deviation is obtained by averaging. .

[0074] in, This represents the current generation number, used to describe the current search progress of the genetic algorithm, and its value ranges from 1 to... ; This represents the maximum number of generations, used to limit the maximum number of iterations in the genetic algorithm; it can be set to 100. Indicates the first The average expected water level deviation of the population is used to reflect the water improvement capability of the current overall irrigation scheme for the population.

[0075] (2) Calculate the first The expected deviation of the average moisture content of the first generation population from the second generation population The difference in the expected average water content deviation between generations of the population yields the rate of deterioration of water content deviation. , Indicates the first The rate of deterioration of the expected deviation of average moisture content in the population.

[0076] Specifically, if A value greater than 0 indicates that the current population's candidate water and fertilizer programs have a weakened ability to improve water conditions, suggesting a deteriorating trend in water status; if... A value less than or equal to 0 indicates that the current population's candidate water and fertilizer programs have not worsened their ability to improve water conditions.

[0077] Then, the rate of deterioration of moisture deviation and the threshold of the rate of deterioration of moisture deviation are calculated. The ratio of these values ​​is truncated to non-negative values ​​to obtain the deterioration intensity adjustment amount. Among them, The threshold for the rate of deterioration of moisture deviation is obtained from the daily fluctuation range of moisture deviation under normal management conditions. It is used to distinguish between natural fluctuations and abnormal deterioration and can be taken as 0.08.

[0078] (3) Update the crossover probability based on the evolutionary process and the severity of the deterioration. , Indicates the first The crossover probability in a genetic algorithm is used to control the probability of gene recombination between parent individuals.

[0079] Specifically, the crossover probability is updated based on the evolutionary process and the deterioration intensity adjustment, consisting of an evolutionary decay term and a deterioration suppression term, and its boundary is constrained by a limiting operator. The update method is expressed as follows: in, This represents the initial crossover probability, used to provide the recombination strength in the initial stage of the genetic algorithm, and can be set to 0.85. This represents the crossover probability decay coefficient, used to gradually reduce the intensity of crossover operations as evolution progresses; it can be set to 0.6. This represents the inhibition coefficient of water deviation deterioration on the crossover probability, used to reduce local recombination dependence when water conditions deteriorate rapidly, and can be set to 0.35; This represents the lower bound of the crossover probability, used to avoid the crossover operation from completely failing; it can be set to 0.45. This represents the upper limit of the crossover probability, used to avoid excessive crossover from destroying the structure of superior individuals; it can be set to 0.95. This represents the minimum value function, used to limit the maximum influence of the deterioration adjustment term; This represents the maximum value function, used to limit the deterioration adjustment term to be non-negative, ensuring that suppression is triggered only when the deviation worsens; This represents the natural exponential function, used to describe the continuous adjustment of crossover probability during the evolutionary process; This represents the limiting operator, used to restrict the calculation result to a certain range. to between.

[0080] The update method for the crossover probability is mainly determined by two parts: The term is used to ensure that the crossover probability decays exponentially with the number of generations, thus achieving a smooth transition from the early global search to the later local fine search. The term is an adaptive regulator that can further reduce the crossover probability when the water deviation is detected to worsen, reduce the dependence on the current local combination, and suppress excessive gene recombination in regions that may be ineffective at present.

[0081] (4) Update the mutation probability according to the evolutionary process and the degree of deterioration adjustment. , Indicates the first The mutation probability of a generational genetic algorithm is used to control the probability of random perturbations in offspring individuals.

[0082] In one implementation, the mutation probability is updated based on the evolutionary process and the deterioration intensity adjustment, consisting of an evolutionary enhancement term and a deterioration enhancement term, and its boundary is constrained by a limiting operator. The update method is expressed as follows: in, This represents the initial mutation probability, used to provide the perturbation strength in the initial stage of the genetic algorithm, and can be set to 0.04. This represents the mutation probability enhancement coefficient, used to maintain the necessary perturbation ability as evolution progresses, and can be taken as 0.3; This represents the enhancement coefficient of the mutation probability due to the deterioration of water conditions. It is used to expand the scope of new candidate water and fertilizer solutions when the water condition deteriorates rapidly. It can be taken as 1.5. This represents the lower bound of the mutation probability, used to maintain the minimum perturbation capability, and can be set to 0.01; This represents the upper limit of the mutation probability, used to avoid excessive mutation leading to random search; a value of 0.25 is acceptable. This represents the maximum value function, used to limit the deterioration adjustment term to be non-negative, ensuring that enhancement is triggered only when the deviation worsens; This represents the natural exponential function, used to describe the continuous adjustment of the probability of mutation during the evolutionary process; This represents the limiting operator, used to restrict the calculation result to a certain range. to between.

[0083] The way mutation probabilities are updated is mainly determined by two parts: The term is used to ensure that the probability of mutation increases exponentially with the number of generations, so as to maintain population diversity in the later stages of evolution and prevent premature convergence. The term is an adaptive regulator that can significantly increase the mutation probability when it detects a deterioration in water deviation, allowing the algorithm to jump out of the current search area and explore new, potentially more robust water and fertilizer solutions.

[0084] As an example, if the average expected deviation of water content for two consecutive generations of candidate water and fertilizer schemes increases from 0.42 to 0.55, and the threshold for the rate of deterioration of water content deviation is set to 0.08, the system determines that the current water status has a significant deterioration trend. The crossover probability is lowered, the mutation probability is raised, and the genetic algorithm explores new combinations of irrigation duration and fertilizer ratio. If the average expected deviation of water content decreases from 0.42 to 0.36, the deterioration adjustment term is not triggered, and the genetic algorithm is more inclined to retain and refine existing excellent schemes through crossover operations.

[0085] 6. Chaotic-Reverse Elite Initialization and Polynomial Mutation Offspring Generation Since the water and fertilizer decision variables include single irrigation duration, single nitrogen fertilizer application rate, single phosphorus fertilizer application rate, and single potassium fertilizer application rate, and these variables have different dimensions, boundaries, and obvious coupling relationships, random initialization can easily lead to the initial population concentrating in local areas. Excessive water and fertilizer candidate solutions can also increase the number of ineffective fitness evaluations. This step uses chaotic mapping to enhance the coverage of the initial population and combines it with reverse candidate solutions to expand the distribution range of candidate solutions. Fitness ranking is used to select elite initial populations. During the generational evolution process, simulated binary crossover, polynomial mutation, and elite retention mechanisms are used to generate offspring, thus balancing global search capability and the inheritance of superior solutions. The specific process is as follows: Figure 7 As shown, a detailed description follows: (1) Set the boundary of water and fertilizer decision variables.

[0086] Specifically, the first The lower bound of each decision variable is denoted as . The upper boundary is recorded as .

[0087] in, This represents the index of water and fertilizer decision variables, with values ​​ranging from 1 to 4. The first water and fertilizer decision variable is the duration of a single irrigation, the second is the amount of nitrogen fertilizer applied in a single irrigation, the third is the amount of phosphorus fertilizer applied in a single irrigation, and the fourth is the amount of potassium fertilizer applied in a single irrigation. The lower bound of each water and fertilizer decision variable can be set to 0, and the upper bound of each water and fertilizer decision variable can be set to 90 minutes, 45 kg / ha, 30 kg / ha, and 60 kg / ha, respectively.

[0088] (2) Generate based on Logistic chaotic mapping A chaotic candidate solution.

[0089] First, generate initial chaotic variables in the range of 0 to 1 for each candidate solution and each water and fertilizer decision variable; then, iteratively update them using a Logistic chaotic mapping, with the chaotic mapping coefficient set to 3.99; finally, map the chaotic variables to their actual values ​​according to the upper and lower bounds of the corresponding water and fertilizer decision variables to obtain the chaotic candidate solutions. , Indicates the first There are chaotic candidate solutions, with dimension . .in, This represents the candidate solution index, with a value ranging from 1 to... ; This represents the population size, used to control the number of candidate water and fertilizer schemes retained in each generation of the algorithm; a value of 60 is acceptable.

[0090] The update method for the Logistic chaotic mapping is as follows: In the formula, Indicates the first The chaotic variable in the next iteration has a value range between 0 and 1; Indicates the first The chaotic variable in the next iteration has a value range between 0 and 1; Let be the chaotic mapping coefficients, when When the system is in a completely chaotic state, the mapping can produce widely distributed and non-repeating sequences.

[0091] (3) Construct an inverse candidate solution for each chaotic candidate solution. , Indicates the first There are inverse candidate solutions, with dimension . It is used to supplement the search area not covered by chaotic candidate solutions.

[0092] Specifically, for the first For each water and fertilizer decision variable, a reverse candidate value is constructed based on its upper and lower bounds, meaning the reverse candidate value is located symmetrically to the original candidate value about the center of the interval. If the rate of deterioration of water deviation in the previous optimization cycle exceeds the water deviation deterioration rate threshold, the reverse candidate solution is proportionally reduced towards the low input direction. The reduction ratio can be 10% to avoid excessive concentration of the initial population in the high water and fertilizer input area when the root zone state deteriorates rapidly or the sensor fluctuates abnormally. If the rate of deterioration of water deviation in the previous optimization cycle does not exceed the water deviation deterioration rate threshold, a reverse candidate solution is generated according to the standard upper and lower bounds to ensure the coverage of the initial search space.

[0093] As an example, if the boundary of a single irrigation duration is... In a chaotic candidate solution, the irrigation duration is 30 minutes. Its symmetric point about the center of the interval (45 minutes) is... The irrigation duration component in this reverse candidate solution is 60 minutes. If the rate of deterioration of water deviation in the previous optimization cycle exceeds the threshold at this time, the reverse candidate solution needs to be reduced by 10% towards lower input, meaning the final reverse value becomes... minute.

[0094] (4) A number of chaotic candidate solutions and The reverse candidate solutions are merged to obtain the initial candidate solution set. , This represents the initial set of candidate solutions, which includes chaotic candidate solutions and reverse candidate solutions, and is used to expand the initial search range.

[0095] Then, each candidate solution in the initial candidate solution set is substituted into the fitness function. The evaluation was conducted, and the fitness function values ​​were sorted from smallest to largest, with the top performers selected. The candidate solutions form the initial population. , This represents the initial population, used as the starting point for the evolutionary search in the genetic algorithm.

[0096] (5) In the generational evolution stage, the current parent population is selected according to the fitness value, and the candidate water and fertilizer decision vectors with smaller fitness values ​​are preferentially retained to participate in crossover and mutation operations to generate offspring.

[0097] Specifically, the selection operation can adopt a tournament selection method, that is, each time 3 to 5 candidate water and fertilizer decision vectors are randomly selected from the population, and the candidate water and fertilizer decision vector with the smallest fitness value is selected as the parent individual, so as to balance the inheritance of excellent individuals and population diversity.

[0098] (6) Based on the crossover probability Perform a simulated binary cross.

[0099] First, for the paired selected parent candidate water and fertilizer decision vectors, a... Uniformly distributed random numbers within the interval Secondly, use this random number. With cross probability If a comparison is made, Less than If so, then perform continuous variable crossover operation on each water and fertilizer decision variable of the parent vector; otherwise, do not perform crossover, and directly retain the parent vector as the child vector. Then, if Less than Then, simulated binary crossover is performed, and the two parent individuals are in the first... The values ​​of the decision variables are denoted as follows: and and generate a Uniformly distributed random numbers Calculate the first Expansion coefficients of water and fertilizer decision variables The calculation method is expressed as follows: in, To simulate the distribution index of binary crossover, a larger distribution index value results in offspring variable values ​​that are closer to their parent variable values. A value of 15 is suitable for local combinatorial searches within a continuous water and fertilizer decision space. Finally, two offspring variable values ​​are generated. and , This indicates that the first offspring individual after crossover is in the [number]th [position]. The values ​​of each water and fertilizer decision variable This indicates that the second offspring individual after crossover is in the [number]th generation. The values ​​of the water and fertilizer decision variables are calculated as follows: (7) Based on the probability of mutation Perform polynomial mutation.

[0100] First, for each water and fertilizer decision variable in each offspring candidate water and fertilizer decision vector, generate a... Uniformly distributed random numbers within the interval Secondly, use this random number. With the probability of mutation If a comparison is made, Less than If so, a new variable value is generated based on the upper and lower bounds of the variable and the multinomial distribution disturbance; otherwise, the variable value remains unchanged. Then, if Less than Then, a polynomial mutation is performed, and the offspring individual on the The values ​​of the decision variables are denoted as follows: Its boundary is and generate a Uniformly distributed random numbers Calculate the first Disturbance factors of water and fertilizer decision variables The calculation method is expressed as follows: in, The distribution index of the multinomial variation can make small perturbations have a higher probability and medium or large perturbations have a lower probability. It can be set to 20, which can refine existing good water and fertilizer programs and explore new irrigation and fertilization combinations when the deviation worsens.

[0101] Finally, the new variable values ​​after mutation are generated. , This indicates that the offspring individual after the mutation is in the [number]th [year]. The new values ​​for the water and fertilizer decision variables are: Furthermore, boundary repair is performed on water and fertilizer decision variables that exceed the boundaries after crossover and mutation. If a variable value is below the lower bound, it is corrected to the corresponding lower bound; if a variable value is above the upper bound, it is corrected to the corresponding upper bound. Boundary repair ensures that all candidate water and fertilizer decision vectors satisfy the constraints of irrigation equipment capacity, fertilizer applicator capacity, pipeline injection capacity, and local fertilization procedures.

[0102] (8) Adopting elite retention Mechanism updates the population.

[0103] in, This refers to the size of the parent population. This refers to the size of the offspring population produced through selection, crossover, and mutation. The corresponding parental population size is , The corresponding offspring population size is also .

[0104] In the specific implementation, first include The parent population of each individual and The offspring populations of each individual merge to form a population containing A mixed population of individuals is generated, and then all individuals in the mixed population are substituted into the fitness function to calculate their fitness values. These individuals are then sorted in ascending order of fitness values, and the individuals at the top of the list are selected. The optimal individual is selected as the next generation population. Based on this, it is possible to avoid the loss of discovered excellent water and fertilizer combinations during random crossover or random mutation, while retaining newly generated candidate water and fertilizer schemes to enter the competition.

[0105] (9) Determine whether the genetic algorithm meets the termination condition. If the condition is met, stop the evolution and output the optimal water and fertilizer decision vector corresponding to the individual with the smallest fitness value in the current population. .

[0106] Specifically, if the improvement in the optimal fitness value over 10 consecutive generations is less than If the genetic algorithm reaches a stable state, it is considered to have stopped evolving; if the current generation reaches the maximum number of generations, it also stops evolving.

[0107] Output the optimal water and fertilizer decision vector after termination. , can be represented as: in, This represents the optimal water and fertilizer decision vector, used to send execution parameters to the irrigation controller and fertilizer controller; Indicates the optimal duration for a single irrigation session; This indicates the optimal amount of nitrogen fertilizer to apply in a single application. This indicates the optimal amount of phosphate fertilizer to be applied in a single application. This indicates the optimal amount of potassium fertilizer to apply in a single application.

[0108] As an example, if the population size is set to 60, the system first generates 60 chaotic candidate solutions through Logistic chaotic mapping, then generates 60 reverse candidate solutions through reverse construction, merging them to obtain an initial set of 120 candidate solutions; substituting these 120 candidate solutions into the fitness function for calculation, if one of the candidate solutions is... The maximum nitrogen-potassium application ratio is taken as 1.2, and the current deviation vector is... The water-nutrient deviation cost was calculated to be 0.35 using the unit water and fertilizer input response coefficient matrix and adaptive weights, and the nitrogen-potassium ratio was [value missing]. If the ratio of nitrogen to potassium application is lower than the maximum, there is no penalty, and the total input cost is 0.1, then the calculated fitness function value is 0.45. After calculating the fitness of all 120 solutions, they are sorted in ascending order of value, and the top 60 optimal solutions are selected as the initial population, which includes the solutions with the fitness function value of 0.45. If, when the genetic algorithm evolves to the 46th generation, the improvement in the optimal fitness function value over 10 consecutive generations is less than... If the algorithm converges and stops evolving, the optimal water and fertilizer decision vector is output as follows: The system will send the application rates of 12 kg / ha of nitrogen fertilizer, 8 kg / ha of phosphate fertilizer, and 50 kg / ha of potassium fertilizer to the irrigation controller every 48 minutes, and then the irrigation system and the fertilization system will complete the corresponding water and fertilizer operations.

[0109] Analyze the line graph of the fitness convergence process of the genetic algorithm, such as... Figure 8 As shown, the trend of fitness value with the number of generations during the search process of the improved genetic algorithm is illustrated. The horizontal axis represents the number of generations (dimensionless), and the vertical axis represents the fitness value (dimensionless). The red solid line in the figure represents the optimal fitness value, and the blue dashed line represents the average fitness value of the population. The experiment shows that the optimal fitness value decreases rapidly in the first 20 generations and then tends to stabilize. The average fitness value decreases overall with the number of generations and the fluctuation gradually decreases.

[0110] Analyze the fitness cost decomposition histogram of candidate water and fertilizer programs, such as... Figure 9 As shown, the graph illustrates the composition of various costs in the fitness function for multiple candidate water and fertilizer schemes. The horizontal axis represents the candidate scheme name (dimensionless), and the vertical axis represents the fitness function value (dimensionless). Different colored bars in the graph represent the costs of water-nutrient deviation, nitrogen-potassium antagonism, irrigation input, and fertilizer input, respectively, with the total fitness value (dimensionless) marked at the top of each bar. The experiment shows that different schemes differ significantly in deviation improvement, antagonism risk, and economic cost. Scheme 4 has the lowest total fitness value, demonstrating that the fitness function can comprehensively weigh agronomic effects, nutrient synergy risks, and resource input, effectively selecting the optimal water and fertilizer scheme.

[0111] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not 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 or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for synergistic optimization of water and fertilizer management in taro based on digital IoT data acquisition, characterized in that, include: S1. Obtain multidimensional data of taro root zone, calculate the instantaneous change amplitude of adjacent points to mark pulse intervals, divide irrigation pulse intervals and non-pulse stable intervals, perform data processing on the observation indicators in the irrigation pulse intervals and non-pulse stable intervals to obtain irrigation pulse response values ​​and stable scale transformation values, and then generate event-driven scale transformation values ​​and agronomic state correction values. S2. Determine the current growth stage of taro, and call the optimal target values ​​of water and nutrients corresponding to the current growth stage and the normal relative fluctuation coefficient to calculate the standardized deviation of the agronomic state correction value, so as to obtain the water-nutrient coupling deviation characteristics. S3. Using soil available potassium concentration as a reference sequence for potassium absorption status in the root zone, and soil nitrate nitrogen concentration and soil available phosphorus concentration as comparison sequences, a nitrogen-phosphorus-potassium antagonism strength factor was constructed based on sliding grey relational analysis. S4. An improved genetic algorithm is used to find the optimal water and fertilizer decision vector. The improvements to the genetic algorithm include: a fitness function that integrates deviation, antagonism factor and input penalty, adaptive crossover mutation probability adjustment based on deviation deterioration rate, chaotic-reverse initialization and elite retention mechanism to update the population.

2. The method for synergistic optimization of water and fertilizer management for taro based on digital IoT data acquisition as described in claim 1, characterized in that, S1 specifically includes: Acquire multidimensional raw observation sequences of taro root zone, irrigation valve opening signals, and fertilizer pump operation signals; First-order difference calculations were performed on adjacent sampling points of the taro root zone observation index to obtain the instantaneous change amplitude of the index. Pulse interval markings were obtained based on irrigation valve opening signals, fertilizer pump operation signals, and instantaneous change amplitudes. Based on the pulse interval marking, the irrigation pulse interval and non-pulse stationary interval are divided. The event benchmark centering process is performed on the taro root area observation index in the irrigation pulse interval to obtain the irrigation pulse response value. The exponential moving average and minimum-maximum normalization process is performed on the taro root area observation index in the non-pulse stationary interval to obtain the stationary scale transformation value. Event-driven scale transformation values ​​and agronomic state correction values ​​are generated based on pulse interval markings.

3. The method for synergistic optimization of water and fertilizer management for taro based on digital IoT data acquisition as described in claim 1, characterized in that, S2 specifically includes: The number of days the taro grows is calculated based on the planting date and the current date, and the growth period markers of the taro are determined in conjunction with field phenological records; Based on the taro growth stage identifier, the water and fertilizer target parameter table is called to obtain the optimal target value and normal relative fluctuation coefficient of the water-nutrient index in the current growth stage; The agronomical state correction values ​​of the water-nutrient index are compared with the optimal target values ​​for the current growth period. The relative deviation is calculated first, and then divided by the normal relative fluctuation coefficient to obtain the standardized deviation. The standardized deviations corresponding to soil volumetric water content, soil nitrate nitrogen concentration, soil available phosphorus concentration, and soil available potassium concentration are arranged in a fixed order to form a water-nutrient coupling deviation vector.

4. The method for synergistic optimization of water and fertilizer management for taro based on digital IoT data acquisition as described in claim 1, characterized in that, S3 specifically includes: A sliding grey relational time window was selected, and soil nitrate nitrogen concentration sequence, soil available phosphorus concentration sequence, and soil available potassium concentration sequence were extracted within the sliding grey relational time window and then normalized. Using the normalized soil available potassium concentration sequence as a reference sequence, the point-by-point differences between the normalized soil nitrate nitrogen concentration sequence and the soil available potassium concentration sequence, as well as the point-by-point differences between the normalized soil available phosphorus concentration sequence and the soil available potassium concentration sequence, were calculated to obtain the nitrogen-potassium correlation coefficient sequence and the phosphorus-potassium correlation coefficient sequence, respectively. The nitrogen-potassium correlation coefficient sequence and the phosphorus-potassium correlation coefficient sequence are averaged to obtain the nitrogen-potassium grey correlation degree and the phosphorus-potassium grey correlation degree, respectively. The nitrogen-phosphorus-potassium antagonism strength factor is calculated based on the nitrogen-potassium grey relational degree and the phosphorus-potassium grey relational degree.

5. The method for synergistic optimization of water and fertilizer management for taro based on digital IoT data acquisition as described in claim 1, characterized in that, The fitness function in S4 specifically includes: Candidate water and fertilizer decision vectors are constructed based on single irrigation duration, single nitrogen fertilizer application rate, single phosphate fertilizer application rate, and single potash fertilizer application rate. Based on historical irrigation test data, the unit water and fertilizer input response coefficient matrix is ​​pre-calibrated, and a unit water and fertilizer input response model is constructed based on the unit water and fertilizer input response coefficient matrix. Based on the candidate water and fertilizer decision vectors, the expected deviation vector after application is obtained using a response model; Adaptive importance weights are determined based on the current growth stage of taro and recent deviations. Estimate the expected nitrogen, phosphorus, and potassium antagonistic strength factor based on the current nitrogen, phosphorus, and potassium antagonistic strength factor and candidate water and fertilizer decision vectors; A fitness function is constructed by integrating expected deviation, nitrogen-potassium ratio penalty, irrigation input penalty, and fertilizer input penalty.

6. The method for synergistic optimization of water and fertilizer management for taro based on digital IoT data acquisition as described in claim 1, characterized in that, The S4 crossover mutation probability adjustment specifically includes: The rate of deterioration of water deviation is the difference between the average expected water deviation of two adjacent generations of population. The average expected moisture deviation is obtained by averaging the expected soil volume moisture deviations corresponding to all candidate water and fertilizer decision vectors in the statistical population. The ratio of the rate of deterioration of moisture deviation to the threshold of the rate of deterioration of moisture deviation is calculated and truncated to a non-negative value to obtain the deterioration intensity adjustment amount; The crossover and mutation probabilities are updated based on the evolutionary process and the degree of deterioration.

7. The method for synergistic optimization of water and fertilizer management for taro based on digital IoT data acquisition as described in claim 1, characterized in that, S4 Chaos - Reverse Initialization specifically includes: Set the boundaries of water and fertilizer decision variables, and generate initial chaotic variables in the range of 0 to 1 for each candidate solution and each water and fertilizer decision variable; Chaotic mapping iterative update is adopted to map the chaotic variables to their actual values ​​according to the upper and lower bounds of the corresponding water and fertilizer decision variables, thereby obtaining chaotic candidate solutions; For each chaotic candidate solution, construct an inverse candidate solution. Then merge the chaotic candidate solutions and the inverse candidate solutions to obtain the initial candidate solution set.

8. The method for synergistic optimization of water and fertilizer management for taro based on digital IoT data acquisition as described in claim 1, characterized in that, The S4 elite retention mechanism includes: Simulated binary crossover is performed based on the crossover probability, and polynomial mutation is performed based on the mutation probability. An elite preservation mechanism is used to update the population, merging the parent population and the offspring population after crossover mutation to form a mixed population; In the mixed population, the fitness values ​​of all individuals are calculated and sorted from smallest to largest. The half of the individuals with the smaller fitness values ​​are taken as the next generation population. The algorithm determines whether to proceed to the next iteration or output the final result based on the stopping condition.

9. A method for synergistic optimization of water and fertilizer management for taro based on digital IoT data acquisition, as described in claim 8, is characterized in that... The algorithm's stopping conditions include: The algorithm stops when the improvement in the optimal fitness value is less than the set value for 10 consecutive generations or when the number of iterations reaches the maximum value. The vector corresponding to the individual with the smallest fitness value in the current population is output as the optimal water and fertilizer decision vector. Otherwise, the next round of the loop is performed.

Images