A pipeline irrigation constant pressure water supply device and an intelligent control method thereof

Abstract

The present application relates to a kind of pipeline irrigation constant pressure water supply device and its intelligent control method, including pipeline irrigation constant pressure water supply device and constant pressure intelligent control method.Pipeline irrigation constant pressure water supply device contains irrigation module, filter module, photovoltaic module and control module.Intelligent constant pressure control method realizes the constant pressure control of irrigation pipeline by the following steps: first, the real-time operation data of pipeline irrigation water supply device is obtained, then data preprocessing is carried out and sample set is constructed;Intelligent prediction model (3P3Z-IDF) coupling three zero three pole compensation and adaptive deep forest is proposed, and the effectiveness of model is verified and regulated in real time.The technology improves the accuracy and response speed of control system, and enhances the adaptive ability of system, can automatically adjust water pressure according to environmental changes and real-time data.

Description

Technical Field

[0001] This invention relates to a constant pressure water supply device for pipeline irrigation and its intelligent control method, belonging to the field of agricultural automation technology. Background Technology

[0002] Pipeline irrigation is an irrigation engineering technology that uses pipelines instead of open channels to transport water, and it has become one of the most promising new irrigation methods in modern agriculture. However, the stability of water pressure in the pipelines can be affected by factors such as pipeline deformation, blockage, and aging, resulting in low stability and utilization of irrigation water within the pipeline irrigation area, often leading to uneven irrigation in localized areas. Therefore, ensuring stable water pressure in the pipelines is a fundamental prerequisite for achieving stable irrigation water supply.

[0003] Existing constant-pressure irrigation systems typically maintain stable water pressure through a combination of sensors, controllers, pumps, and PID control algorithms. However, this traditional constant-pressure control method has several drawbacks, such as requiring manual labor, being prone to errors, having low accuracy, slow response speed, and weak anti-interference capabilities. Therefore, there is an urgent need to develop a novel constant-pressure water supply device for pipeline irrigation and its intelligent control method to achieve intelligent, real-time regulation of water pressure in the irrigation area, further improving the efficiency and stability of pipeline irrigation. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a constant pressure water supply device for pipeline irrigation and its intelligent control method.

[0005] To achieve the above-mentioned objectives, the present invention adopts the following technical solution: a pipeline irrigation constant pressure water supply device, characterized in that it includes an irrigation module, a filtration module, a photovoltaic module and a control module, wherein the irrigation module, the filtration module, the photovoltaic module and the control module are all installed on a mounting frame;

[0006] The irrigation module includes an inlet, a main water supply pipe, a main water pump, a standby pump, a pump base, a regulating tank, a regulating tank inlet pipe, a regulating tank inlet solenoid valve, a regulating tank outlet pipe, a regulating tank outlet solenoid valve, an irrigation water delivery pipe, an operating characteristic sensor, and an environmental sensor.

[0007] The system comprises: a regulating tank fixed on a mounting frame; a regulating tank inlet pipe and a regulating tank outlet pipe connected to both ends of the regulating tank, both of which are internal to the regulating tank; a main water supply pipe with one end connected to an inlet and the other end connected to the regulating tank outlet pipe, with a regulating tank outlet solenoid valve installed on the outlet pipe; a regulating tank inlet pipe with one end connected to the regulating tank and the other end connected to the side wall of the main water supply pipe, also internal to the main water supply pipe, with a regulating tank inlet solenoid valve installed on the inlet pipe; a performance sensor for monitoring flow and pressure data within the main water supply pipe, located between the filter module and the regulating tank inlet pipe; and a main water pump and a standby pump both fixed on the mounting frame, with the main water pump connected to both the main water supply pipe and the irrigation water supply pipe, and the standby pump connected to both the main water supply pipe and the irrigation water supply pipe.

[0008] The filtration module is located after the water inlet, which is connected to the main water supply pipe through the filtration module. The filtration module includes a sand filter and a disc filter. The water inlet, sand filter, disc filter, and main water supply pipe are connected in sequence. The sand filter is equipped with a sludge discharge port and a pipeline solenoid valve. The sludge discharge port is connected to the front end of the sand filter, and the pipeline solenoid valve is connected to the constant pressure regulating controller in the control box to control the opening and closing of the main water supply pipe. The sludge discharge port is equipped with a sludge discharge solenoid valve, which is connected to the constant pressure regulating controller to control the opening and closing of the sludge discharge port. When the sludge discharge solenoid valve is closed and the pipeline solenoid valve is opened, water can pass through the sand filter to filter out large particles, and then through the disc filter for fine filtration. When too much sediment accumulates on the sand filter and the disc filter, the sludge discharge solenoid valve is opened and the pipeline solenoid valve is closed to flush the sand filter and the disc filter.

[0009] The control module includes a control box and a power manager. The control box is mounted on a mounting bracket, and the power manager is mounted behind the control box and fixed to the mounting bracket. The control box contains a photovoltaic controller and a constant voltage regulator controller, and the power manager is responsible for regulating the power supply balance between the photovoltaic system and the mains power.

[0010] The photovoltaic module includes a photovoltaic panel, a photovoltaic inverter, a battery cooling circulator, battery cooling circulation pipes, and a battery pack. The photovoltaic panel is mounted on top of a mounting frame, and the photovoltaic inverter is fixed on the mounting frame and located behind the control box. The battery pack is located between the main water pump and the standby pump, and a battery cooling circulator is installed at its rear top, connected to the bottom of the regulating tank via cooling circulation pipes. The photovoltaic panel is connected to the main water pump and the battery pack via a panel inverter and a power manager. The photovoltaic panel prioritizes power supply to the main water pump, and excess power is stored in the battery pack. When the battery pack overheats, the battery cooling circulator increases the flow rate in the cooling circulation pipes. An environmental sensor is installed above the battery pack.

[0011] When the power in the photovoltaic module is insufficient to maintain the normal operation of the main water pump, the power manager automatically switches to mains power supply.

[0012] The main water pump is connected to the operating characteristic sensor via a constant pressure regulating controller in the control box. The operating characteristic sensor is responsible for monitoring the pressure and flow data in the irrigation module. The constant pressure regulating controller is connected to the inlet solenoid valve and outlet solenoid valve of the regulating tank, and adjusts the speed of the main water pump according to the data from the operating characteristic sensor. When the operating characteristic sensor detects that the water pressure in the main water supply pipe is too high, the constant pressure regulating controller will reduce the speed of the main water pump; when the water pressure is too low, it will increase the speed. If the water pressure is too high and the main water pump adjustment is ineffective, the constant pressure regulating controller will open the inlet solenoid valve of the regulating tank to balance the water pressure; if the water pressure is too low and the adjustment is ineffective, the constant pressure regulating controller will open the outlet solenoid valve of the regulating tank to balance the water pressure.

[0013] The regulating tank is fixed to the mounting frame by a regulating tank bracket. A cleaning port is located on the top of the regulating tank, covered by a cleaning port cap. A water level sensor is installed on the regulating tank. The water level sensor is connected to the regulating tank inlet solenoid valve, the regulating tank outlet solenoid valve, and the pipeline solenoid valve via a constant pressure regulating controller in the control box. When the water level sensor detects that the water level has reached the preset upper limit, it sends a signal to the constant pressure regulating controller to open the regulating tank outlet solenoid valve and close the regulating tank inlet solenoid valve and the pipeline solenoid valve. When the water level drops to the preset intermediate water level, the water level sensor sends a signal to the constant pressure regulating controller to close the regulating tank outlet solenoid valve and open the pipeline solenoid valve. When the water level sensor detects that the water level is below the preset lower limit, it sends a signal to the constant pressure regulating controller to open the regulating tank inlet solenoid valve and close the outlet solenoid valve. After the water level rises again to the preset intermediate water level, constant pressure irrigation continues.

[0014] Both the main water pump and the standby pump are equipped with a water pump base at the bottom, which is fixed on the mounting frame. Both the main water pump and the standby pump are fixed on the mounting frame through the water pump base.

[0015] In the irrigation module, the main water pump is connected to the operating characteristic sensor via a constant pressure regulating controller in the control box. The operating characteristic sensor monitors the pressure and flow data in the irrigation module. The constant pressure regulating controller is connected to the inlet and outlet solenoid valves of the regulating tank and adjusts the speed of the main water pump based on the data obtained by the operating characteristic sensor. When the operating characteristic sensor detects that the water pressure in the main water supply pipe is too high, it sends a signal to the constant pressure regulating controller to reduce the speed of the main water pump. When the operating characteristic sensor detects that the water pressure in the main water supply pipe is too low, it sends a signal to the constant pressure regulating controller to increase the speed of the main water pump. When the operating characteristic sensor detects that the water pressure in the main water supply pipe is too high, and adjusting the main water pump cannot relieve the water pressure, the constant pressure regulating controller in the control box opens the inlet solenoid valve of the regulating tank to balance the water pressure. When the operating characteristic sensor detects that the water pressure in the main water supply pipe is too low, and adjusting the flow rate of the main water pump cannot relieve the water pressure, it is necessary to increase the pressure in the main water supply pipe, and the constant pressure regulating controller opens the outlet solenoid valve of the regulating tank to balance the water pressure.

[0016] The filtration module is located at the front end of the inlet pipe of the regulating tank. After the irrigation water is drawn in from the inlet, the solenoid valve at the sludge discharge port is closed and the solenoid valve in the pipeline is opened. The irrigation water will first pass through the sand and gravel filter to filter out large debris such as sand and gravel, and then pass through the disc filter for fine filtration. When too much sediment accumulates in the sand and gravel filter and the disc filter, the solenoid valve at the sludge discharge port is opened and the solenoid valve in the pipeline is closed. Under the action of water flow, the inner wall of the pipeline is flushed and most of the sediment is carried away from the sludge discharge port.

[0017] In the photovoltaic module, the photovoltaic inverter is connected to the main water pump and battery pack via a power manager. The photovoltaic panel prioritizes powering the main water pump, while the redundant power is stored in the battery pack. When the light intensity is insufficient, the photovoltaic controller in the control box will use the battery pack as the power source. A battery cooling circulator is installed behind the battery pack and is connected to the regulating tank via a battery cooling circulating water pipe. The battery pack has a temperature monitoring function. When the battery pack temperature is detected to be too high, the battery cooling circulating pump controller sends a command to the battery cooling circulating pump to increase the flow rate, accelerating the cooling circulation in the battery cooling circulating water pipe.

[0018] The control module contains a photovoltaic controller and a constant voltage regulator controller, while the power manager is used to balance the power supply between the photovoltaic system and the mains power. When the power from the photovoltaic module is insufficient to maintain the normal operation of the main water pump, the power manager will automatically switch the power supply to the mains power.

[0019] When using this system, first turn on the power manager switch. At this time, the constant pressure regulating controller located in the control box receives the irrigation command and sends a start signal to the main water pump. Based on real-time operating data provided by the operating characteristic sensor, the main water pump automatically adjusts its speed to ensure a constant water supply pressure. During irrigation, irrigation water is drawn in through the inlet, purified by a sand filter and a disc filter, then flows into the main water supply pipe, and is pumped by the main water pump to the irrigation delivery pipe, ultimately reaching the crops to complete the irrigation.

[0020] The system also features an intelligent monitoring function. When the operating characteristic sensor detects an abnormality or malfunction of the main water pump, the constant pressure regulating controller in the control box will automatically switch the water pump system to the standby water pump to ensure that the water supply pressure remains constant throughout the irrigation process.

[0021] A constant-pressure intelligent control method for a pipeline irrigation constant-pressure water supply device is proposed. This method employs a coupled three-zero-point three-pole compensation method and an adaptive deep forest intelligent prediction model (3P3Z-IDF) to perform real-time constant-pressure control of the pipeline irrigation water supply device. The specific steps are as follows:

[0022] S1. Obtain real-time operating data of the pipeline irrigation constant pressure water supply device: Obtain real-time operating characteristic data of the main water pump through the operating characteristic sensor installed at the inlet of the pipeline irrigation constant pressure water supply device; obtain real-time operating environment data of the pipeline irrigation constant pressure water supply device through the environmental sensor installed above the battery pack; obtain real-time control data of the main water pump through the constant pressure regulating controller installed in the control box;

[0023] S2. Data Preprocessing: Data cleaning algorithms are used to clean the real-time operating data of the pipeline irrigation constant pressure water supply device obtained in S1;

[0024] S3. Constructing a sample set: The real-time operation data of the pipeline irrigation constant pressure water supply device collected after cleaning in S2 is split into 80% as the training set and 20% as the test set, and data augmentation methods are used to strengthen the training set.

[0025] S4. Creating a Constant Pressure Control Strategy: A smart prediction model 3P3Z-IDF coupled with three zero-point and three-pole compensation and adaptive deep forest is proposed. By adjusting the parameters of the poles and zeros in real time, constant pressure control of the pipeline irrigation constant pressure water supply device is realized. The specific method is as follows:

[0026] S4-1. Set a desired water pressure value u s , which serves as the control pressure for the main water pump; set the initial water pressure u(0). u(0) is the minimum output of the main water pump to ensure that the system has a stable initial state during startup;

[0027] S4-2. At any time k, measure the current water pressure u of the main water pump. a and control pressure u s The error e(k) is obtained by comparison; then the error increment Δe(k) is calculated based on two adjacent errors, and the calculation method is as follows:

[0028] Δe(k)=e(k)-e(k-1) (1)

[0029] In the formula, e(k-1) and e(k) are the errors between the water pressure at time k-1 and k and the expected water pressure, and Δe(k) is the increment of the water pressure error between adjacent times;

[0030] S4-3. Determine the technical parameters of the zeros and poles based on the real-time verified 3P3Z-IDF intelligent prediction model;

[0031] S4-4. Based on the determined technical parameters of the zero and pole points, calculate the control components of the main water pump, whose transfer function is:

[0032]

[0033] In the formula, K is the gain coefficient; T Z1 T Z2 T Z3 The zeros in the transfer function; T P1 T P2 T P3 The poles are the transfer function, and s is a complex variable in the Laplace transform.

[0034] Discretize the transfer function G(s) to obtain the control quantity of the main water pump. The calculation method is as follows:

[0035]

[0036] In the formula, Δu(k) is the control increment at time k, which is the change in the controller output; Δe(k) is the error increment at time k, which is the change in the difference between the expected value and the actual value; Z0, Z1, Z2, and Z3 are zero-point control coefficients, used to adjust the influence of the error increment on the control increment; and P1, P2, and P3 are pole control coefficients, used to adjust the influence of past control increments on the current control increment.

[0037] S4-5. Based on the calculated control increment Δu(k), adjust the output speed of the main water pump in real time to change the water pressure; repeat steps S4-3 to S4-5 in real time.

[0038] In S1, the operating characteristic sensor acquires real-time data on the main water pump's operating characteristics, including: pump speed, flow rate, and actual pressure; the environmental sensor acquires real-time data on the operating environment of the constant pressure water supply device, including: temperature and humidity in the device; the real-time control data of the water pump acquired by the constant pressure regulating controller in the control box includes: input current, input voltage, control pressure, deviation between control pressure and actual pressure, zero and pole control parameters, and output control quantity.

[0039] In S2, a data cleaning algorithm is used to clean the real-time operating data of the pipeline irrigation constant pressure water supply device obtained in S1. The specific calculation method is as follows:

[0040]

[0041]

[0042]

[0043] In the formula, data points A and B are data points obtained in S1; d(A,B) is the distance between points A and B; N k (A) represents the k nearest neighbors of point A; w(A,B) is the weight between points A and B; σ is the median distance between data points A and B; WLRD k (A) and WLRD k (B) is the reachability density of data point A and data point B; WLOF k (A) is the outlier score of data point A;

[0044] When the WLOF of data point A k (A) Data points with values ​​higher than 95% or lower than 5% are considered abnormal data and need to be removed.

[0045] In S3, samples are randomly selected from the training set, and augmented samples are generated for each sample in the following manner. These augmented samples are then added to the training set. The calculation method for the augmented samples is as follows:

[0046] X' i =X i +λ·sgn(L adv (X i ,Y i ))+r(θ) (7)

[0047]

[0048] In the formula, X i For training samples; X' i To enhance the sample; Y i For sample X iThe true value; For sample X i The predicted value; λ is the perturbation value; sgn() is the sign function, used to randomly set a positive or negative sign; L adv (X i ,Y i ) is the loss function; r(θ) is a random repeating vector.

[0049] In S4-3, the 3P3Z-IDF intelligent prediction model is trained through a feature scanning module and a training module. The specific steps are as follows:

[0050] The S5-1 feature scanning module creates windows of different sizes to perform sliding scans on the original training set, learns its local features, and then concatenates the learned local features to obtain the final feature vector.

[0051] The S5-2 training module takes the feature vectors obtained in S5-1 and inputs them into this module for training. The input feature vectors are processed by the ensemble learners of each layer to generate class vectors. The class vectors generated by each layer are formed by concatenating the multi-dimensional class vectors generated by all learners in that layer with the original feature vectors, and are used as the input for the next layer. This process continues until the last layer becomes the final output of the model. When the accuracy of the current layer does not improve compared to the previous layer, the construction of the next layer is stopped, and the optimal model is obtained by adjusting the model parameters.

[0052]

[0053] X l+1 =[X,O l (10)

[0054] in, This involves concatenating the results of the K learners in this layer; h l i (X l ) is the feature vector X input to the current layer by the i-th learner in layer l. l The processing result; O l X is the combination of the outputs of all learners in layer l; X is the original feature vector; X l+1 It is the feature vector of layer l+1;

[0055] Simultaneously, the hyperparameters in the ensemble learner are used as individual vectors of the population for differential evolution. The specific steps are as follows:

[0056] ① Hyperparameter population initialization: Generate hyperparameters randomly in the search space and initialize them as the initial hyperparameter population;

[0057] ② Individual mutation; mutated individuals are generated according to the DE / currentto-pbest-λ mutation strategy; the mutation method of DE / currentto-pbest-λ is as follows:

[0058] V i,g =x i,g +F i ·λ·(x p_best,g -x i,g )+F i ·λ·(x r1,g -x r2,g (11)

[0059] In the formula, V i,g Let x be the test vector at position i; i,g Let x be the group vector at position i; p_best,g For the individual with the best fitness in the population, p∈(0,1); F i For x r1,g and x r2,g The associated mutation factor; λ is an adaptive scaling factor. If a better solution is found in several generations, the value of λ is gradually increased until an upper limit is reached. If no improvement is found in several generations, λ is gradually decreased. The initial value of λ is 0.5; x r1,g and x r2,g Two different individuals were randomly selected;

[0060] ③ Crossover operation; improve algorithm flexibility by adding an adaptive machine; adjust CR in each generation of the population based on the success rates of mutation and crossover, respectively. i and F i Two control parameters; if F i If F ≥ 1, then the cutoff value is 1. i If the value is ≤0, then it is regenerated; the formula is as follows:

[0061] CR i =randn i (μ CR ,0.1) (12)

[0062] μ CR = (1-c)·μ CR +c·mean A (S CR (13)

[0063] F i =randc i (μ F ,0.1) (14)

[0064] μ F = (1-c)·μ F+c·mean L (S F (15)

[0065] In the formula, randn i Random operators that are normally distributed, randc i For the random operator of the Cauchy distribution, mean A The arithmetic mean; mean L Lehmer mean, S CR It is a set of all successful cross-possibilities, S F It is the set of all successfully mutated factors. μ CR and μ F S CR and S F The movement center is initialized to 0.5; c is an empirical constant.

[0066] ④ Individual selection; The fitness of individuals is calculated by comparing total entropy, and the individual with the lowest total entropy is selected to enter the next generation of the population;

[0067]

[0068]

[0069] In the formula, N(D) L ) and N(D R ) represents the number of data in the left and right child nodes, E(D) is the entropy of dataset D, C(D) is the set of unique classes in dataset D, and p(k) is the probability of class k in dataset D;

[0070] ⑤ Update and output the results; the algorithm terminates when the population meets the fitness requirement; otherwise, return to step ② to continue the evolutionary cycle.

[0071] In S4-3, real-time validation is performed. After training, the model's effectiveness is verified by calculating the prediction error skewness and prediction error kurtosis based on the validation set in S3. After model deployment, the model's prediction results and actual results are continuously monitored and collected, along with real-time operating data of the pipeline irrigation constant pressure water supply device. Evaluation and validation units are used in 60-minute intervals, and the prediction error skewness and prediction error kurtosis are calculated. The calculation method is as follows:

[0072] e k =y k -y′ k (18)

[0073]

[0074]

[0075]

[0076] In the formula, e k Let y be the sample prediction error of the k-th test sample. k For the actual parameters in the k-th test sample, y' k Here, S represents the prediction parameters corresponding to the k-th test sample, S represents the prediction error skewness, K represents the prediction error kurtosis, n represents the number of test samples in the test sample set, and e represents the prediction parameters. k Let n be the sample prediction error for the k-th test sample, and n be the number of samples.

[0077] If the prediction error skewness and / or prediction error kurtosis are between [-1, 1], it can be determined that the prediction errors of multiple test samples conform to a normal distribution, that is, the reliability of the trained 3P3Z-IDF prediction model is good; if the prediction error skewness and / or prediction error kurtosis are not between [-1, 1], the existing model needs to be used as the initial 3P3Z-IDF model for retraining until the model prediction error skewness and / or prediction error kurtosis are between [-1, 1].

[0078] This invention discloses a constant-pressure water supply device for pipeline irrigation and its intelligent control method, comprising: a constant-pressure water supply device for pipeline irrigation and a constant-pressure intelligent control method. The constant-pressure water supply device for pipeline irrigation includes an irrigation module, a filtration module, a photovoltaic module, and a control module. The intelligent constant-pressure control method achieves constant-pressure control of the irrigation pipeline through the following steps: first, acquiring real-time operating data of the pipeline irrigation water supply device; then, performing data preprocessing and constructing a sample set; proposing an intelligent prediction model (3P3Z-IDF) coupled with three-zero-point and three-pole compensation and adaptive deep forest, and realizing real-time verification and adjustment of the model's effectiveness. This technology improves the accuracy and response speed of the control system, while enhancing the system's adaptive capability, enabling it to automatically adjust water pressure according to environmental changes and real-time data. Attached Figure Description

[0079] Figure 1 This is an overall schematic diagram of the pipeline irrigation constant pressure water supply device provided in the embodiments of this application;

[0080] Figure 2 This is a schematic diagram of the irrigation module of the pipeline irrigation constant pressure water supply device provided in the embodiments of this application;

[0081] Figure 3 This is a schematic diagram of the filter module of the pipeline irrigation constant pressure water supply device provided in the embodiments of this application;

[0082] Figure 4 This is a schematic diagram of the photovoltaic module and control module of the pipeline irrigation constant pressure water supply device provided in the embodiments of this application;

[0083] Figure 5This is a schematic diagram of the left side of the pipeline irrigation constant pressure water supply device provided in the embodiments of this application;

[0084] Figure 6 This is a schematic diagram of the photovoltaic panel of the pipeline irrigation constant pressure water supply device provided in the embodiments of this application;

[0085] Figure 7 This is a flowchart of the intelligent prediction model of coupled three zero-point and three pole compensation and adaptive deep forest (3P3Z-IDF) provided in the embodiments of this application;

[0086] Figure 8 This is a flowchart of the deep forest decision-making process provided in the embodiments of this application;

[0087] Figure 9 This is a schematic diagram illustrating the response speed of the 3P3Z-IDF intelligent prediction model and PID algorithm provided in the embodiments of this application;

[0088] In the diagram: 1-1 Inlet, 1-2 Main water supply pipe, 1-3 Main water pump, 1-4 Standby pump, 1-5 Pump base, 1-6 Regulating tank, 1-7 Regulating tank inlet pipe, 1-8 Regulating tank inlet solenoid valve, 1-9 Regulating tank outlet pipe, 1-10 Regulating tank outlet solenoid valve, 1-11 Irrigation water supply pipe, 1-12 Cleaning port, 1-13 Cleaning port top cover, 1-14 Operating characteristic sensor, 1-15 Water level sensor, 1-16 Environmental sensor, 1-17 Regulating tank bracket, 2-1 Sand and gravel filter, 2-2 Disc filter, 2-3 Sludge discharge port, 2-4 Sludge discharge port solenoid valve, 2-5 Pipeline solenoid valve, 3-1 Photovoltaic panel, 3-2 Photovoltaic inverter, 3-3 Battery cooling circulator, 3-4 Battery cooling circulation pipeline, 3-5 Battery pack, 4-1 Control box, 4-2 Power manager. Detailed Implementation

[0089] The present invention will be further described below with reference to the accompanying drawings and description.

[0090] A pipeline irrigation constant pressure water supply device, the device as follows Figure 1 , 2 As shown in Figures 3 and 4, the system includes an irrigation module, a filter module 2, a photovoltaic module 3, and a control module 4, all of which are installed in a unified mounting bracket 5.

[0091] Irrigation module 1 includes the following components: inlet 1-1, main water supply pipe 1-2, main water pump 1-3, standby pump 1-4, pump base 1-5, regulating tank 1-6, regulating tank inlet pipe 1-7, regulating tank inlet solenoid valve 1-8, regulating tank outlet pipe 1-9, regulating tank outlet solenoid valve 1-10, irrigation water delivery pipe 1-11, cleaning port 1-12, cleaning port top cover 1-13, operating characteristic sensor 1-14, water level sensor 1-15, environmental sensor 1-16, and regulating tank bracket 1-17. The main water supply pipe 1-2 is equipped with operating characteristic sensor 1-14 for monitoring flow and pressure data. The regulating tank 1-6 is connected to the main water supply pipe 1-2 via the regulating tank inlet pipe 1-7 and the regulating tank outlet pipe 1-9, and is equipped with inlet solenoid valve 1-8 and outlet solenoid valve 1-10. The regulating tank 1-6 is fixed to the mounting bracket via bracket 1-17. The top of the regulating tank 1-6 is equipped with a cleaning port 1-12 and a cleaning port cover 1-13, and a water level sensor 1-15 is installed nearby. The main water pump 1-3 and the standby pump 1-4 are both fixed to the frame via pump bases 1-5 and are respectively connected to the main water supply pipe and the irrigation water delivery pipe. An environmental sensor 1-16 is installed above the battery pack 3-5 between the main water pump and the standby pump.

[0092] The main water pump 1-3 is connected to the operating characteristic sensor 1-14 via a constant pressure regulating controller within the control box 4-1. This sensor monitors the pressure and flow data in the irrigation module. The constant pressure regulating controller adjusts the speed of the main water pump based on the sensor data. When the sensor detects excessively high water pressure in the main water supply pipe, the controller reduces the main water pump speed; when the water pressure is too low, it increases the speed. If the water pressure is too high and the main water pump regulation is ineffective, the inlet solenoid valve 1-8 of the regulating tank will be opened to balance the water pressure; if the water pressure is too low and regulation is ineffective, the outlet solenoid valve 1-10 of the regulating tank will be opened to balance the water pressure.

[0093] Filter module 2 includes a sand filter 2-1, a disc filter 2-2, a sludge discharge port 2-3, a sludge discharge port solenoid valve 2-4, and a pipeline solenoid valve 2-5. This module is located after the inlet 1-1. First, there is the sand filter 2-1, with a sludge discharge port 2-3 at its front end, equipped with the sludge discharge port solenoid valve 2-4. Before the sludge discharge port, the pipeline solenoid valve 2-5 is located. After water is drawn in, the sludge discharge port solenoid valve 2-4 is closed first, and the pipeline solenoid valve 2-5 is opened. The water passes through the sand filter 2-1 to filter large particles, and then through the disc filter 2-2 for fine filtration. When too much sediment accumulates in the sludge discharge port 2-3, the sludge discharge port solenoid valve 2-4 is opened, and the pipeline solenoid valve 2-5 is closed for flushing.

[0094] The photovoltaic module 3 consists of a photovoltaic panel 3-1, a photovoltaic inverter 3-2, a battery cooling circulator 3-3, a battery cooling circulation pipe 3-4, and a battery pack 3-5. The photovoltaic panel 3-1 is mounted on top of the mounting frame, and the photovoltaic inverter 3-2 is fixed to the rear of the control box 4-1 and mounted on the mounting frame 4. The battery pack 3-5 is located between the main water pump 1-3 and the standby pump 1-4, with the battery cooling circulator 3-3 mounted on its rear top, connected to the bottom 1-6 of the regulating tank 3-4 via a cooling circulation water pipe. The photovoltaic panel 3-1 is connected to the main water pump 1-3 and the battery pack 3-5 via the panel inverter 3-2 and the power manager 4-2. The photovoltaic panel 3-1 prioritizes power supply to the main water pump 1-3, with excess power stored in the battery pack 3-5. When the battery pack 3-5 overheats, the battery cooling circulator 3-3 automatically increases the flow rate in the cooling circulation pipe 3-4.

[0095] Control module 4 includes a control box 4-1 and a power manager 4-2. The control box 4-1 is mounted on the edge of the mounting bracket 5, and the power manager 4-2 is mounted behind the control box 4-1 and fixed to the mounting bracket 5. The control box 4-1 contains a photovoltaic controller and a constant voltage regulator controller. The power manager 4-2 is responsible for regulating the power supply balance between the photovoltaic module 3 and the mains power. When the power from the photovoltaic module 3 is insufficient to maintain the normal operation of the main water pump 1-3, the power manager automatically switches to mains power supply.

[0096] When using the system, first turn on the power manager 4-2. At this time, the constant pressure regulating controller located in the control box 4-1 receives the irrigation command and sends a start signal to the main water pump 1-3. The main water pump 1-3 automatically adjusts its speed based on real-time operating data provided by the operating characteristic sensor 1-14 to ensure a constant water supply pressure. During irrigation, irrigation water is drawn in from the inlet 1-1, purified by the sand filter 2-1 and the disc filter 2-2, then flows into the main water supply pipe 1-2, and is delivered by the main water pump 1-3 to the irrigation water delivery pipe 1-11, finally reaching the crops to complete the irrigation.

[0097] The system also features an intelligent monitoring function. When the operating characteristic sensor 1-14 detects an abnormality or malfunction of the main water pump 1-3, the constant pressure regulating controller in the control box 4-1 will automatically switch the water pump system to the standby water pump 1-4 to ensure that the water supply pressure remains constant during the irrigation process.

[0098] This invention also discloses a smart prediction algorithm (3P3Z-IDF) for pipeline irrigation water supply device constant pressure control, which proposes a three-zero-point three-pole compensation method and adaptive deep forest, including the following steps:

[0099] (S1) Obtain real-time operating data of the pipeline irrigation water supply device. In this application, a total of 356 sets of real-time operating data of the pipeline irrigation water supply device are collected, including real-time data of pump operating characteristics, real-time data of pump operating environment, and real-time control data of pump; among them, the real-time data of pump operating characteristics can be collected by operating characteristic sensors, including: pump speed, flow rate, and actual pressure; the real-time data of pump operating environment can be collected by environmental sensors, including: temperature, humidity, etc.; the real-time control data of pump can be collected by the constant pressure regulating controller in the control box, including: input current, input voltage, control pressure, deviation between control pressure and actual pressure, zero and pole control parameters, and output control quantity.

[0100] (S2) Data Preprocessing. In this application, to address potential noise and outliers in the real-time operating data of the pipeline irrigation water supply device collected in S1, the data can be cleaned using the following method. The specific calculation method is as follows:

[0101]

[0102]

[0103]

[0104] In the formula, data points A and B are data points obtained in S1; d(A,B) is the distance between points A and B; N k (A) represents the k nearest neighbors of point A; w(A,B) is the weight between points A and B; σ is the median distance between data points A and B; WLRD k (A) and WLRD k (B) is the reachability density of data point A and data point B; WLOF k (A) is the outlier score of data point A.

[0105] When the WLOF of data point A k (A) Data points with values ​​higher than 95% or lower than 5% are considered outliers and are removed.

[0106] (S3) Constructing a sample set. In this application, the sample set includes multiple samples, each sample set including real-time operating data of the pipeline irrigation water supply device after S2 cleaning. The collected data is divided for training and testing, with 80% used for training (284 groups in total) and 20% for testing (72 groups in total). To enable the 3P3Z-IDF model to better handle noise and outliers and have better robustness and generalization ability, samples are randomly selected from the training set, and augmented samples are generated for each sample in the following manner, and these augmented samples are added to the training set. The calculation method for the augmented samples is as follows:

[0107] X' i =X i +λ·sgn(L adv (X i ,Y i ))+r(θ) (25)

[0108]

[0109] In the formula, X i For training samples; X' i To enhance the sample; Y i For sample X i The true value; For sample X i The predicted value; λ is the perturbation value; sgn() is the sign function, used to randomly set a positive or negative sign; L adv (X i ,Y i ) is the loss function; r(θ) is a random repeating vector.

[0110] (S4). Creating a constant-voltage control strategy. This application proposes an intelligent prediction model (3P3Z-IDF) coupled with three-zero-three-pole compensation and adaptive deep forest. By controlling the poles and zeros, the system achieves better dynamic characteristics, thus providing more precise control. The control strategy is shown below:

[0111] S4-1. Set a desired water pressure value u s The initial water pressure is set to u(0), which is the control pressure of the water pump. u(0) is the minimum output of the water pump to ensure that the system has a stable initial state when it starts up.

[0112] S4-2. At any time k, measure the current water pressure value u. a and control water pressure u s The error e(k) is obtained by comparison. Then, the error increment Δe(k) is calculated based on two adjacent errors, and the calculation method is as follows:

[0113]

[0114] In the formula, e(k-1) and e(k) are the errors between the water pressure at time k-1 and k and the expected water pressure, and Δe(k) is the increment of the water pressure error between adjacent times.

[0115] S4-3. Determine the technical parameters of zeros and poles based on the real-time verified 3P3Z-IDF intelligent prediction model.

[0116] S4-4. Based on the determined technical parameters of the zero point and poles, calculate the control components of the water pump, whose transfer function is:

[0117]

[0118] In the formula, K is the gain coefficient; T Z1 T Z2 T Z3 The zeros in the transfer function; T P1 T P2 T P3 Let be the poles in the transfer function, and s be the complex variable in the Laplace transform.

[0119] Discretize the transfer function G(s) to obtain the control quantity of the water pump. The calculation method is as follows:

[0120]

[0121] In the formula, Δu(k) is the control increment at time k, which is the change in the controller output; Δe(k) is the error increment at time k, which is the change in the difference between the expected value and the actual value; Z0, Z1, Z2, and Z3 are zero-point control coefficients, used to adjust the influence of the error increment on the control increment; and P1, P2, and P3 are pole control coefficients, used to adjust the influence of past control increments on the current control increment.

[0122] S4-5. Based on the calculated control increment Δu(k), adjust the output speed of the main water pump (1-3) in real time to change the water pressure. Repeat steps S4-3 to S4-5 in real time.

[0123] (S5). The 3P3Z-IDF intelligent prediction model of S4-3 is trained through the feature scanning module and the training module. The specific steps are as follows:

[0124] The S5-1 feature scanning module creates windows of different sizes to perform a sliding scan on the original training set, learning its local features. The learned local features are then concatenated to obtain the final feature vector.

[0125] The S5-2 training module takes the feature vectors obtained from S5-1 and inputs them into this module for training. The input feature vectors are processed by the ensemble learners in each layer to generate class vectors. Each class vector generated by a layer is constructed by concatenating the multi-dimensional class vectors generated by all learners in that layer with the original feature vectors, and this concatenation serves as the input for the next layer. This process continues layer by layer until the last layer becomes the final output of the model. If the accuracy of the current layer does not improve compared to the previous layer, the construction of the next layer stops, and the optimal model is obtained by adjusting the model parameters.

[0126]

[0127] X l+1 =[X,Ol (31)

[0128] in, This involves concatenating the results of the K learners in this layer; h l i (X l ) is the feature vector X input to the current layer by the i-th learner in layer l. l The processing result; O l X is the combination of the outputs of all learners in layer l; X is the original feature vector; X l+1 It is the feature vector of layer l+1.

[0129] Simultaneously, the hyperparameters in the ensemble learner are used as individual vectors of the population for differential evolution. The specific steps are as follows:

[0130] ① Hyperparameter population initialization. Hyperparameters are randomly generated in the search space and initialized to the initial hyperparameter population.

[0131] ② Individual Mutation. Mutated individuals are generated according to the DE / currentto-pbest-λ mutation strategy. The mutation method of DE / currentto-pbest-λ is as follows:

[0132] V i,g =x i,g +F i ·λ·(x p_best,g -x i,g )+F i ·λ·(x r1,g -x r2,g (32)

[0133] In the formula, V i,g Let x be the test vector at position i; i,g Let x be the group vector at position i; p_best,g For the individual with the best fitness in the population, p∈(0,1); F i For x r1,g and x r2,g The associated mutation factor; λ is an adaptive scaling factor. If a better solution is found in several generations, the value of λ is gradually increased until an upper limit is reached. If no improvement is found in several generations, λ is gradually decreased. The initial value of λ is 0.5; x r1,g and x r2,g Two different individuals were randomly selected.

[0134] ③ Crossover operation. The algorithm's flexibility is improved by adding an adaptive machine. In each generation of the population, the CR is adjusted based on the success rates of mutation and crossover, respectively. i and F i Two control parameters. If Fi If F ≥ 1, then the cutoff value is 1. i If the value is ≤0, then it is regenerated. The formula is as follows:

[0135] CR i =randn i (μ CR ,0.1) (33)

[0136] μ CR = (1-c)·μ CR +c·mean A (S CR (34)

[0137] F i =randc i (μ F ,0.1) (35)

[0138] μ F = (1-c)·μ F +c·mean L (S F (36)

[0139] In the formula, randn i Random operators that are normally distributed, randc i For the random operator of the Cauchy distribution, mean A The arithmetic mean; mean L Lehmer mean, S CR It is a set of all successful cross-possibilities, S F It is the set of all successfully mutated factors. μ CR and μ F S CR and S F The moving center is initialized to 0.5; c is an empirical constant.

[0140] ④ Individual selection. The fitness of individuals is calculated by comparing total entropy, and the individual with the lowest total entropy is selected to enter the next generation of the population.

[0141]

[0142]

[0143] In the formula, N(D) L ) and N(D R ) represents the number of data in the left and right child nodes, E(D) is the D entropy of the dataset, C(D) is the set of unique classes in the dataset D, and p(k) is the probability of class k in the dataset D.

[0144] ⑤. Result Update and Output. The algorithm terminates when the population meets the fitness requirement. Otherwise, return to step ② to continue the evolutionary cycle.

[0145] (S6) After training is complete, the model's effectiveness is verified by calculating the prediction error skewness and prediction error kurtosis on the validation set based on S3. After model deployment, the model's prediction results and actual results are continuously monitored and collected, with each evaluation and validation unit lasting 60 minutes. The prediction error skewness and prediction error kurtosis are calculated. The calculation method is as follows:

[0146] e k =y k -y′ k (39)

[0147]

[0148]

[0149]

[0150] In the formula, e k Let y be the sample prediction error of the k-th test sample. k For the actual parameters in the k-th test sample, y' k Here, S represents the prediction parameters corresponding to the k-th test sample, S represents the prediction error skewness, K represents the prediction error kurtosis, n represents the number of test samples in the test sample set, and e represents the prediction parameters. k Let n be the sample prediction error of the k-th test sample, and n be the number of samples.

[0151] If the prediction error skewness and / or prediction error kurtosis are between [-1, 1], it can be determined that the prediction errors of multiple test samples conform to a normal distribution, that is, the reliability of the trained 3P3Z-IDF prediction model is good. If the prediction error skewness and / or prediction error kurtosis are not between [-1, 1], the existing model needs to be used as the initial 3P3Z-IDF model for retraining until the model's prediction error skewness and / or prediction error kurtosis are within [-1, 1].

Claims

1. A constant pressure water supply device for pipeline irrigation, characterized in that, It includes an irrigation module (1), a filter module (2), a photovoltaic module (3) and a control module (4), all of which are mounted on a mounting frame; The irrigation module (1) includes an inlet (1-1), a main water supply pipe (1-2), a main water pump (1-3), a standby pump (1-4), a water pump base (1-5), a regulating tank (1-6), a regulating tank inlet pipe (1-7), a regulating tank inlet solenoid valve (1-8), a regulating tank outlet pipe (1-9), a regulating tank outlet solenoid valve (1-10), an irrigation water delivery pipe (1-11), an operating characteristic sensor (1-14), and an environmental sensor (1-16). Wherein: the regulating tank (1-6) is fixed on the mounting bracket, the regulating tank inlet pipe (1-7) and the regulating tank outlet pipe (1-9) are respectively connected to both ends of the regulating tank (1-6), and the regulating tank inlet pipe (1-7) and the regulating tank outlet pipe (1-9) are both connected to the inside of the regulating tank (1-6); one end of the main water supply pipe (1-2) is connected to the inlet (1-1), and the other end is connected to the regulating tank outlet pipe (1-9), and the regulating tank outlet solenoid valve (1-10) is installed on the regulating tank outlet pipe (1-9); one end of the regulating tank inlet pipe (1-7) is connected to the regulating tank (1-6), and the other end is connected to the side wall of the main water supply pipe (1-2), and the regulating tank inlet pipe (1-7) is connected to the inside of the main water supply pipe (1-2), the regulating tank inlet pipe (1-7) is connected to the inside of the main water supply pipe (1-2), the regulating tank inlet pipe (1-7) is connected to the side wall of the main water supply pipe (1-2), and the regulating tank inlet pipe (1-7) is connected to the inside of the main water supply pipe (1-2), the regulating tank inlet pipe (1-7) is connected to the side wall of the main water supply pipe (1-6), and the regulating tank outlet pipe (1-7) is connected to the side wall of the main water supply pipe (1-6), ... A regulating tank inlet solenoid valve (1-8) is installed on 1-7; a cleaning port (1-12) is provided on the top of the regulating tank (1-6), and a cleaning port cover (1-13) is provided on the cleaning port (1-12); an operating characteristic sensor (1-14) for monitoring the flow and pressure data in the water supply main pipe (1-2) is installed on the water supply main pipe (1-2), and the operating characteristic sensor (1-14) is located between the filter module (2) and the regulating tank inlet pipe (1-7); the main water pump (1-3) and the standby pump (1-4) are both fixed on the mounting frame, the main water pump (1-3) is connected to the water supply main pipe (1-2) and the irrigation water delivery pipe (1-11) respectively, and the standby pump (1-4) is connected to the water supply main pipe (1-2) and the irrigation water delivery pipe (1-11) respectively. The filter module (2) is located after the inlet (1-1), and the inlet (1-1) is connected to the main water supply pipe (1-2) through the filter module (2). The filter module (2) includes a sand filter (2-1) and a disc filter (2-2). The inlet (1-1), sand filter (2-1), disc filter (2-2), and main water supply pipe (1-2) are connected in sequence. The sand filter (2-1) is equipped with a sludge discharge port (2-3) and a pipeline solenoid valve (2-5). The sludge discharge port (2-3) is connected to the front end of the sand filter (2-1), and the pipeline solenoid valve (2-5) is connected to the constant pressure regulating controller in the control box (4-1) to control the water supply. The water main pipe (1-2) is opened and closed; the sludge discharge port (2-3) is equipped with a sludge discharge port solenoid valve (2-4), which is connected to a constant pressure regulating controller to control the opening and closing of the sludge discharge port (2-3); when the sludge discharge port solenoid valve (2-4) is closed and the pipeline solenoid valve (2-5) is opened, water can pass through the sand and gravel filter (2-1) to filter large particles, and then through the disc filter (2-2) for fine filtration; when too much sediment accumulates on the sand and gravel filter (2-1) and the disc filter (2-2), the sludge discharge port solenoid valve (2-4) is opened and the pipeline solenoid valve (2-5) is closed, so that the sand and gravel filter (2-1) and the disc filter (2-2) can be flushed; The control module (4) includes a control box (4-1) and a power manager (4-2). The control box (4-1) is mounted on a mounting bracket, and the power manager (4-2) is mounted behind the control box (4-1) and fixed on the mounting bracket. The control box (4-1) contains a photovoltaic controller and a constant voltage regulator controller, and the power manager (4-2) is responsible for regulating the power supply balance between the photovoltaic and the mains power. The photovoltaic module (3) includes a photovoltaic panel (3-1), a photovoltaic inverter (3-2), a battery cooling circulator (3-3), a battery cooling circulation pipeline (3-4), and a battery pack (3-5); the photovoltaic panel (3-1) is mounted on the top of the mounting frame, and the photovoltaic inverter (3-2) is fixed on the mounting frame and located behind the control box (4-1); the battery pack (3-5) is located between the main water pump (1-3) and the standby pump (1-4), and its rear top is equipped with a battery cooling circulator (3-3), which is connected to the regulating tank (1) through the battery cooling circulation pipeline (3-4). -6) Bottom connection; the photovoltaic panel (3-1) is connected to the main water pump (1-3) and the battery pack (3-5) through the photovoltaic inverter (3-2) and power manager (4-2); the photovoltaic panel (3-1) prioritizes power supply to the main water pump (1-3), and the excess power is stored in the battery pack (3-5); the battery pack (3-5) has a temperature monitoring function, and when the battery pack (3-5) overheats, the battery cooling circulator (3-3) increases the flow rate in the battery cooling circulation pipe (3-4); an environmental sensor (1-16) is installed above the battery pack (3-5). When the power in the photovoltaic module (3) is insufficient to maintain the normal operation of the main water pump (1-3), the power manager (4-2) automatically switches to mains power supply; The main water pump (1-3) is connected to the operating characteristic sensor (1-14) through the constant pressure regulating controller in the control box (4-1); the operating characteristic sensor (1-14) is responsible for monitoring the pressure and flow data in the irrigation module (1); the constant pressure regulating controller is connected to the regulating tank inlet solenoid valve (1-8) and the regulating tank outlet solenoid valve (1-10), and regulates the speed of the main water pump (1-3) according to the data of the operating characteristic sensor (1-14); when the operating characteristic sensor (1-14) detects that the water pressure in the main water supply pipe (1-2) is too high, the constant pressure regulating controller will reduce the speed of the main water pump (1-3); when the water pressure is too low, the speed will be increased; if the water pressure is too high and the main water pump (1-3) regulation is ineffective, the constant pressure regulating controller will open the regulating tank inlet solenoid valve (1-8) to balance the water pressure; if the water pressure is too low and the regulation is ineffective, the constant pressure regulating controller will open the regulating tank outlet solenoid valve (1-10) to balance the water pressure. The device employs a coupled three-zero-point three-pole compensation method and an adaptive deep forest intelligent prediction model (3P3Z-IDF) to perform real-time constant pressure control on the water supply device for pipeline irrigation. The specific steps are as follows: S1. Obtain real-time operating data of the pipeline irrigation constant pressure water supply device: Obtain real-time operating characteristic data of the main water pump (1-3) through the operating characteristic sensor (1-14) installed at the inlet (1-1) of the pipeline irrigation constant pressure water supply device; obtain real-time operating environment data of the pipeline irrigation constant pressure water supply device through the environmental sensor (1-16) installed above the battery pack (3-5); obtain real-time control data of the main water pump (1-3) through the constant pressure regulating controller installed in the control box (4-1); S2. Data Preprocessing: Data cleaning algorithms are used to clean the real-time operating data of the pipeline irrigation constant pressure water supply device obtained in S1; S3. Constructing a sample set: The real-time operation data of the pipeline irrigation constant pressure water supply device collected after cleaning in S2 is split into 80% as the training set and 20% as the test set. Data augmentation methods are used to strengthen the training set. S4. Creating a Constant Pressure Control Strategy: A smart prediction model 3P3Z-IDF coupled with three zero-point and three-pole compensation and adaptive deep forest is proposed. By adjusting the parameters of the poles and zeros in real time, constant pressure control of the pipeline irrigation constant pressure water supply device is realized. The specific method is as follows: S4-1. Set a desired water pressure value u s , as the control pressure of the main water pump (1-3); set the initial water pressure u(0); u(0) is the minimum output of the main water pump (1-3) to ensure that the system has a stable initial state when it starts up; S4-2. At any time k, measure the current water pressure u of the main water pump (1-3). a and control pressure u s The error e(k) is obtained by comparison; then the error increment Δe(k) is calculated based on two adjacent errors, and the calculation method is as follows: （1） In the formula, e(k-1) and e(k) are the errors between the water pressure at time k-1 and k and the expected water pressure, and Δe(k) is the increment of the water pressure error between adjacent times; S4-3. Determine the technical parameters of the zeros and poles based on the real-time verified 3P3Z-IDF intelligent prediction model; S4-4. Based on the determined technical parameters of the zero and pole points, calculate the control components of the main water pump (1-3), whose transfer function is: （2） In the formula, K is the gain coefficient; T Z1 T Z2 T Z3 The zeros in the transfer function; T P1 T P2 T P3 The poles are the transfer function, and s is a complex variable in the Laplace transform. Discretize the transfer function G(s) to obtain the control quantity of the main water pump (1-3). The calculation method is as follows: （3） In the formula, Δu(k) is the control increment at time k, which is the change in the controller output; Δe(k) is the error increment at time k, which is the change in the difference between the expected value and the actual value; Z0, Z1, Z2, and Z3 are zero-point control coefficients, used to adjust the influence of the error increment on the control increment; and P1, P2, and P3 are pole control coefficients, used to adjust the influence of past control increments on the current control increment. S4-5. Based on the calculated control increment Δu(k), adjust the output speed of the main water pump (1-3) in real time to change the water pressure; repeat steps S4-3 to S4-5 in real time.

2. The pipeline irrigation constant pressure water supply device according to claim 1, characterized in that, The regulating tank (1-6) is fixed to the mounting frame by the regulating tank bracket (1-17). A water level sensor (1-15) is installed on the regulating tank (1-6). The water level sensor (1-15) is connected to the regulating tank inlet solenoid valve (1-8), the regulating tank outlet solenoid valve (1-10), and the pipeline solenoid valve (2-5) through the constant pressure regulating controller in the control box (4-1). When the water level sensor (1-15) detects that the water level has reached the preset upper limit, the water level sensor (1-15) will send a signal to the constant pressure regulating controller to open the regulating tank outlet solenoid valve (1-10) and close the regulating tank inlet solenoid valve. Water solenoid valve (1-8) and pipeline solenoid valve (2-5); when the water level drops to the preset intermediate water level, the water level sensor (1-15) will send a signal to the constant pressure regulating controller to close the regulating tank outlet solenoid valve (1-10) and open the pipeline solenoid valve (2-5); when the water level sensor (1-15) detects that the water level is lower than the preset lower limit, the water level sensor (1-15) will send a signal to the constant pressure regulating controller to open the regulating tank inlet solenoid valve (1-8) and close the regulating tank outlet solenoid valve (1-10); after the water level rises again to the preset intermediate water level, constant pressure irrigation will continue.

3. The pipeline irrigation constant pressure water supply device according to claim 1, characterized in that, The main water pump (1-3) and the standby pump (1-4) are both equipped with a water pump base (1-5) at the bottom. The base is fixed on the mounting frame. The main water pump (1-3) and the standby pump (1-4) are both fixed on the mounting frame through the water pump base.

4. A pipeline irrigation constant pressure water supply device according to claim 1, characterized in that, In the photovoltaic module (3), the photovoltaic inverter (3-2) is connected to the main water pump (1-3) and the battery pack (3-5) through the power manager (4-2). The photovoltaic panel (3-1) prioritizes power supply to the main water pump (1-3), and the redundant power is stored in the battery pack (3-5). When the light intensity is insufficient, the photovoltaic controller in the control box (4-1) will use the battery pack (3-5) as the power source. The battery pack (3-5) is equipped with a battery cooling circulator (3-3) and is connected to the regulating tank (1-6) through the battery cooling circulation pipeline (3-4). When the temperature of the battery pack (3-5) is detected to be too high, the battery cooling circulator (3-3) sends an instruction to the battery cooling circulation pump to increase the flow rate, thereby accelerating the cooling circulation in the battery cooling circulation pipeline (3-4). In the control module (4), the control box (4-1) contains a photovoltaic controller and a constant voltage regulator controller. The power manager (4-2) is used to regulate the balance between photovoltaic and mains power. When the power in the photovoltaic module (3) is insufficient to maintain the normal operation of the main water pump (1-3), the power manager (4-2) will automatically switch the power to mains power. The system also has an intelligent monitoring function. When the operating characteristic sensor (1-14) detects an abnormality in the main water pump (1-3), the constant pressure regulating controller in the control box (4-1) will automatically switch the water pump system to the standby pump (1-4) to ensure that the water supply pressure remains constant during the irrigation process. When using the system, first turn on the power manager (4-2). The constant pressure regulating controller in the control box (4-1) receives the irrigation command and sends a start signal to the main water pump (1-3). The main water pump (1-3) automatically adjusts its speed according to the real-time operating data provided by the operating characteristic sensor (1-14) to ensure a constant water supply pressure. During the irrigation process, irrigation water is drawn in from the inlet (1-1), then purified through the sand filter (2-1) and the disc filter (2-2), and then flows into the main water supply pipe (1-2), and is sent to the irrigation water delivery pipe (1-11) by the main water pump (1-3), finally reaching the crops to complete the irrigation.

5. A pipeline irrigation constant pressure water supply device according to claim 1, characterized in that, In S1, the operating characteristic sensor (1-14) acquires real-time operating characteristic data of the main water pump (1-3), including: pump speed, flow rate and actual pressure; the environmental sensor (1-16) acquires real-time operating environment data of the irrigation constant pressure water supply device, including: temperature and humidity in the device; the real-time control data of the water pump acquired by the constant pressure regulating controller in the control box (4-1) includes: input current, input voltage, control pressure, deviation between control pressure and actual pressure, zero and pole control parameters and output control quantity.

6. A pipeline irrigation constant pressure water supply device according to claim 1, characterized in that, In S2, a data cleaning algorithm is used to clean the real-time operating data of the pipeline irrigation constant pressure water supply device obtained in S1. The specific calculation method is as follows: （4） （5） （6） In the formula, data points A and B are data points obtained in S1; d(A,B) is the distance between points A and B; N k (A) represents the k nearest neighbors of point A; w(A,B) is the weight between points A and B; σ is the median distance between data points A and B; WLRD k (A) and WLRD k (B) is the reachability density of data point A and data point B; WLOF k (A) is the outlier score of data point A; When the WLOF of data point A k (A) Data points with values ​​higher than 95% or lower than 5% are considered abnormal data and need to be removed.

7. A pipeline irrigation constant pressure water supply device according to claim 1, characterized in that, In S3, samples are randomly selected from the training set, and augmented samples are generated for each sample in the following manner. These augmented samples are then added to the training set. The calculation method for the augmented samples is as follows: （7） （8） In the formula, X i For training samples; X' i To enhance the sample; Y i For sample X i The true value; For sample X i The predicted value; λ is the disturbance value; L is a sign function used to randomly set a positive or negative sign. adv (X i , Y i ) is the loss function; r(θ) is a random repeating vector.

8. A pipeline irrigation constant pressure water supply device according to claim 1, characterized in that, In S4-3, the 3P3Z-IDF intelligent prediction model is trained through a feature scanning module and a training module. The specific steps are as follows: The S5-1 feature scanning module creates windows of different sizes to perform sliding scans on the original training set, learns its local features, and then concatenates the learned local features to obtain the final feature vector. The S5-2 training module takes the feature vectors obtained in S5-1 and inputs them into this module for training. The input feature vectors are processed by the ensemble learners of each layer to generate class vectors. The class vectors generated by each layer are formed by concatenating the multi-dimensional class vectors generated by all learners in that layer with the original feature vectors, and are used as the input for the next layer. This process continues until the last layer becomes the final output of the model. When the accuracy of the current layer does not improve compared to the previous layer, the construction of the next layer is stopped, and the optimal model is obtained by adjusting the model parameters. （9） （10） in, This involves concatenating the results of the K learners in this layer; h l i (X l ) is the feature vector X input to the current layer by the i-th learner in layer l. l The processing result; O l X is the combination of the outputs of all learners in layer l; X is the original feature vector; X l+1 It is the feature vector of layer l+1; Simultaneously, the hyperparameters in the ensemble learner are used as individual vectors of the population for differential evolution. The specific steps are as follows: ① Hyperparameter population initialization: Generate hyperparameters randomly in the search space and initialize them to the initial hyperparameter population; ② Individual mutation; mutated individuals are generated according to the DE / currentto-pbest-λ mutation strategy; the mutation method of DE / currentto-pbest-λ is as follows: （11） In the formula, V i,g Let x be the test vector at position i; i,g Let x be the group vector at position i; p_best,g The individual with the best fitness in the population, p∈(0,1); F i For x r1,g and x r2,g The associated mutation factor; λ is an adaptive scaling factor. If a better solution is found in several generations, the value of λ is gradually increased until an upper limit is reached. If no improvement is found in several generations, λ is gradually decreased. The initial value of λ is 0.5; x r1,g and x r2,g Two different individuals were randomly selected; ③ Crossover operation; improve algorithm flexibility by adding an adaptive machine; adjust CR in each generation of the population based on the success rates of mutation and crossover, respectively. i and F i Two control parameters; if F i If F ≥ 1, then the cutoff value is 1. i If ≤ 0, then regenerate; the formula is as follows: （12） （13） （14） （15） In the formula, randn i Random operators that are normally distributed, randc i For the random operator of the Cauchy distribution, mean A The arithmetic mean; mean L Lehmer mean, S CR It is a set of all successful cross-possibilities, S F It is the set of all successfully mutated factors, μ CR and μ F S CR and S F The movement center is initialized to 0.5; c is an empirical constant. ④ Individual selection; The fitness of individuals is calculated by comparing total entropy, and the individual with the lowest total entropy is selected to enter the next generation of the population; （16） （17） In the formula, N(D) L ) and N(D R ) represents the number of data in the left and right child nodes, E(D) is the entropy of dataset D, C(D) is the set of unique classes in dataset D, and p(k) is the probability of class k in dataset D; ⑤ Update and output the results; the algorithm terminates when the population meets the fitness requirement; otherwise, return to step ② to continue the evolutionary cycle.

9. A pipeline irrigation constant pressure water supply device according to claim 1, characterized in that, In S4-3, real-time validation is performed. After training, the model's effectiveness is verified by calculating the prediction error skewness and prediction error kurtosis based on the validation set in S3. After model deployment, the model's prediction results and real-time operating data of the pipeline irrigation constant pressure water supply device are continuously monitored and collected. Evaluation and validation units are used every 60 minutes, and the prediction error skewness and prediction error kurtosis are calculated. The calculation method is as follows: （18） （19） （20） （21） In the formula, e k Let y be the sample prediction error of the k-th test sample. k For the actual parameters in the k-th test sample, y' k Let S be the prediction parameter corresponding to the k-th test sample, S be the prediction error skewness, K be the prediction error kurtosis, and n be the number of test samples in the test sample set. If the prediction error skewness or prediction error kurtosis is between [-1, 1], it can be determined that the prediction errors of multiple test samples conform to a normal distribution, that is, the reliability of the trained 3P3Z-IDF prediction model is good; if the prediction error skewness or prediction error kurtosis is not between [-1, 1], the existing model needs to be used as the initial 3P3Z-IDF model for retraining until the model prediction error skewness or prediction error kurtosis is between [-1, 1].

Images