Wireless dynamic self-adaptive weighing verification method suitable for poultry breeding

Abstract

The present application relates to the technical field of dynamic weighing, in particular to a wireless dynamic adaptive weighing verification method suitable for poultry breeding, which comprises the following steps: S1, construction of a wireless dynamic adaptive weighing system; S2, automatic skinning algorithm for weight data; S3, preparation before operation of the poultry dynamic weighing algorithm; S4, adaptive weighing of the poultry dynamic weighing algorithm. Through reasonable scheme design and verification of the present application, the present application is compared with the traditional EMD decomposition, the truncated mean method and the wavelet transform algorithm, and practical application monitoring is carried out in the farm, so that a method which simultaneously has speed and accuracy and well meets the requirements of weight detection in the poultry breeding process is obtained, and the method has high practicability. The data collected in the broiler farm show that the weight error is kept within 1% under the condition that the growth of broilers in the cage is uniform, and the weight error is 1 / 3 of the manual weighing error.

Description

Technical Field

[0001] This invention relates to the field of dynamic weighing technology, and specifically to a wireless dynamic adaptive weighing verification method suitable for poultry farming. Background Technology

[0002] Body weight is a crucial indicator for assessing the growth of poultry in poultry farming. It helps farmers determine the poultry's growth status and develop scientific feeding plans. Currently, manual weighing using scales with devices that restrict poultry movement is the primary method for obtaining poultry weight in farming environments. However, this method cannot achieve unrestricted dynamic weighing and suffers from drawbacks such as low worker efficiency, improper operation, and poultry stress. Furthermore, the living habits of poultry can continuously damage the instrument's communication lines, leading to reduced instrument stability and lifespan. Therefore, adopting a wireless networking method to accurately and quickly obtain the true weight of poultry is extremely important.

[0003] Current research on dynamic weighing, both domestically and internationally, primarily focuses on vehicles and mining, with significant studies also conducted on poultry weighing. Zhao Huibing et al. designed an automated, stress-free dynamic weighing system for beef cattle, utilizing the EMD algorithm combined with poultry behavior to acquire weight data during natural walking. However, they did not address issues such as endpoint effects and spurious components encountered during the decomposition process. Chen Kaidong et al. designed a sheep flock-free weight monitoring system, achieving dynamic weighing of sheep based on a random forest algorithm and multiple linear regression. Dong Xiaoning designed a dairy cow dynamic weighing system based on STM32, employing a BP neural network model to measure the dynamic weight of dairy cows. Lü Qiantao et al. improved upon the endpoint effects and spurious components in EMD decomposition and applied this to poultry weighing, but it was not validated in poultry farming. All the dynamic weighing systems designed by these scholars require fencing to restrict livestock activity. Tang Sihao et al. proposed a multi-layer BP neural network based on an adaptive moment estimation optimizer, achieving nonlinear correction of the checkweigher sensor and accurately estimating dynamic weighing results. Peng Y et al. designed a real-time automated system for monitoring the feed intake and weight of chicks in group rearing. This system uses analog circuits and digital filtering to acquire dynamic weight, but it only covers chicks and not the entire rearing cycle. He Z et al. designed a goat herd dynamic weighing system based on the Kalman-EEMD algorithm, achieving high-precision dynamic weighing of individual goats. Chen Chaobo et al. used an RBF neural network to process dynamic weighing data from vehicles at different speeds and verified that the radial basis function network exhibits good speed and accuracy in processing dynamic weighing data.

[0004] Current research on dynamic weighing of live poultry mainly focuses on individual livestock, with little attention paid to farmed poultry. Furthermore, it does not address the issue of tare weight baseline deviation caused by the accumulation of feces and feed during the farming process. Therefore, it is essential to design a dynamic weighing system suitable for poultry farming, addressing requirements such as farmed poultry, stress-free weight acquisition, and automatic tare. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide a wireless dynamic adaptive weighing verification method suitable for poultry farming. Through actual application and monitoring verification in broiler farms, the method can quickly and accurately obtain the weight of poultry, and has good adaptability, stability and robustness.

[0006] The technical solution of this invention is as follows:

[0007] A wireless dynamic adaptive weighing verification method suitable for poultry farming includes the following steps:

[0008] S1. Construction of the Wireless Dynamic Adaptive Weighing System: The system includes feeding cages, which are arranged in a multi-point, spaced manner within the breeding shed. The weighing platform data collector is placed at fixed points within the feeding cages. The weighing platform data collector has a built-in LoRa circuit, which sends the collected weight data to a LoRa-to-RS485 circuit. Several weighing platform data collectors interact with the intelligent gateway through a star-shaped network. The intelligent gateway is used for system data calculation and processing.

[0009] S2. The proposed automatic tare algorithm for weight data: Feces and feed continuously accumulate on the surface of the weighing platform collector. The weight data collected by the system is queued separately according to the first-out rule. Each time data is enqueued, the data in the queue are compared pairwise. When the difference between the pairs of data in the queue is within the range, the data in the queue is filtered by removing extreme values ​​and averaging to obtain the result. The average value is subtracted from the initial value to obtain the difference. A positive value indicates that poultry has been put on the weighing platform collector. Every once in a while, the actual weight of the poultry is obtained by EMD decomposition and used as the new benchmark weight for weighing, thus achieving the automatic tare effect.

[0010] S3. Preparations before running the poultry dynamic weighing algorithm: Before running the algorithm, data cleaning / filtering is performed to remove outliers and during operation, the accuracy requirements are checked to improve the system's operating efficiency.

[0011] S4. Adaptive Weighing Algorithm for Poultry Dynamic Weighing: While continuously collecting data, the weighing platform collector also judges the body shape characteristics of the poultry. When consecutive similar data are detected, and the poultry is in a static state, the median average filtering is applied to this continuous data segment, and the average value is used as the final result. The cleaning / filtering and EMD algorithms are skipped to save system resources. Since boundary effects exist during EMD decomposition, causing errors in the final IMF result, it is necessary to solve the boundary effects before decomposition to reduce system errors. By using mirror extension, the endpoint effects are eliminated while increasing the cardinality of the data, thus improving the accuracy of EMD decomposition. This includes the following steps:

[0012] S41. When the birds are in a non-static state, the data after cleaning is g(t). Find the extreme points in g(t) from head to tail and from tail to head. Use the two extreme points as symmetrical points to perform mirror extension to obtain the data s(t).

[0013] S42. Find all the maxima and minima in s(t) by sorting and comparing, and obtain the upper envelope f by cubic sample interpolation. max (t) and lower envelope f min (t); Take the average of the upper and lower envelopes, denoted as m1(t), that is:

[0014]

[0015] Subtracting the original signal s(t) from m1(t) yields:

[0016] h1(t)=s(t)-m1(t) (2)

[0017] S43. Determine whether h1(t) satisfies the following conditions:

[0018] a) The number of extreme points and zero-crossing points are equal or differ by at most 1;

[0019] b) The mean of the upper and lower envelopes is 0;

[0020] If h1(t) satisfies the above conditions, then h1(t) is the first IMF component, denoted as c1(t); otherwise, h1(t) is used as a new original signal, and the above process is repeated, i.e.:

[0021] h2(t)=h1(t)-m2(t) (3)

[0022] S44. Assume that after repeating k times, the two conditions of the IMF are satisfied. At this point, the first IMF component is obtained, namely:

[0023] c1(t)=h k (t) (4)

[0024] Removing c1(t) from the original signal yields the remaining signal:

[0025] r1(t)=s(t)-c1(t) (5)

[0026] After repeating the above process n times with r1(t) as the new original signal, s(t) is finally decomposed into n IMF components and a residual r representing the trend of the original signal. n (t):

[0027]

[0028] S45. To prevent the EMD from continuously decomposing and causing the derived essential mode functions to lose their physical meaning of instantaneous frequency and instantaneous amplitude, it is necessary to limit the number of signal iterations. The standard deviation criterion is used as the stopping criterion; iteration stops when the standard deviation is between 0.2 and 0.3. The formula is as follows:

[0029]

[0030] During the EMD decomposition process, spurious components may sometimes appear, which are components that have no physical meaning or are unrelated to the signal, thus affecting the accuracy of the final weight result.

[0031] S5. The Pearson correlation coefficient is used to determine whether the current IMF is fraudulent. If the similarity to the original data is below a threshold, it is considered a fraudulent IMF. The correlation coefficient function is as follows:

[0032]

[0033] To remove spurious components, it is necessary to determine the correlation between each IMF component and the original signal. Specifically, the correlation coefficient between each IMF component and the original signal is defined as follows:

[0034]

[0035] Set the threshold to one-tenth of the maximum value in the correlation coefficient sequence, that is:

[0036]

[0037] After removing spurious components using the correlation coefficient method, a one-dimensional array of residual quantities representing the signal trend is obtained. The average value of these residual quantities is calculated using median filtering, and the result is the true weight obtained from this dynamic weighing, denoted as W. Further, in the wireless dynamic adaptive weighing system of step S1, the intelligent gateway uses an STM32F429IGT6 processor, and the weighing platform data acquisition unit uses an AT32F427F8P7 processor. The intelligent gateway integrates a human-machine interface, memory, IoT circuitry, RS484 circuitry, USB circuitry, and a LoRa-to-RS485 converter. The weighing platform data acquisition unit integrates a DC-DC circuit, an AD acquisition circuit, a LoRa circuit, a MOS circuit, and a pressure sensor. The intelligent gateway is connected to at least one weighing platform data acquisition unit via a LoRa circuit.

[0038] Furthermore, in the wireless dynamic adaptive weighing system of step S1, the pressure sensor used by the weighing platform collector is model NA1-20kg, with an accuracy of 2.0mV / V±10%; the AD acquisition circuit uses AD7190, with a maximum noise-free resolution of 22.5 bits and an output rate of 4.7Hz-4.8KHz. The high-precision and high-speed acquisition ensures that the acquisition error of the system is kept within 0.1%.

[0039] Furthermore, the automatic tare algorithm for the weight data in step S2 adopts a queue-based automatic tare algorithm, namely:

[0040] S21. When starting up, record the initial state weight value as z. The weight data collected by the system is separately entered into a queue with a data length of 10 according to the first-out rule. Each time data is enqueued, the data in the queue must be compared pairwise.

[0041] S22. When the difference between any two data points in the queue is less than 5g, perform extreme value removal and averaging filtering on the data in the queue to obtain the result w. Subtract the average value from the initial value to obtain the difference Δ.

[0042] Δ=wz

[0043] Δ is a signed number; a positive value represents an animal stepping onto the scale, and a negative value represents an animal leaving the scale. The value of Δ and the initial weight z are stored. The above steps are then repeated using w as the new initial value. During this process, the following result is obtained: n Each Δ and its corresponding initial weight z n ;

[0044] S23. Every so often, an actual animal weight W is obtained through EMD decomposition. W is compared with Δ. If the difference between the two is within 3g, the corresponding initial weight z is considered to be... iThis is the new initial state, which is used as the new baseline weight for weighing to achieve an automatic tare effect.

[0045] Furthermore, before the poultry dynamic weighing algorithm runs in step S3, an improved RKF-EMD dynamic weighing algorithm is used for data filtering to remove outliers, including the following steps:

[0046] A. Prediction Steps: Assume that k represents the current state of the system, and that the process noise W(k) and system measurement noise V(k) are Gaussian noises with zero mean and covariances Q and R, respectively. The current state of the system can be obtained using Kalman filtering as follows:

[0047] X(k|k-1)=AX(k-1|k-1)+BU(k) (a1)

[0048] Then, from equation (2), the covariance of X(k|k-1) can be obtained as:

[0049] P(k|k-1)=AP(k-1|k-1)A T +Q (a2)

[0050] The prediction steps of Kalman filtering are implemented by equations (2) and (3) to obtain the state of the weighing system in the next time step;

[0051] B. Update steps: First, obtain the Kalman gain Kg(k) as follows:

[0052] Kg(k)=P(k|k-1)H T [HP(k|k-1)H T +R] -1 (b1)

[0053] Outlier removal is achieved through observation residuals, which represent the difference between observed and predicted values. They measure the deviation between observed and predicted data; a larger deviation indicates a higher probability that the current data is an outlier. Let the observation residuals be C(k).

[0054] C(k)=Z(k)-HX(k|k-1) (b2)

[0055] Here, a robustness parameter 'm' is defined to control the sensitivity to outliers. The robustness parameter multiplied by the square root of the state covariance is denoted as E(k), i.e.:

[0056]

[0057] Compare the absolute value of E(k) with the observed residual C(k):

[0058] a) When |C(k)| > E(k), the current data is considered an outlier, and the state prediction value X(k|k - 1) is used to replace the data;

[0059] b) When |C(k)| < E(k), continue to execute the update step, and use the current data to correct the state estimate value;

[0060] When continuing to execute update step B, the prediction result of the current state is obtained by the estimation of equations (2) and (3). Based on the measurement value of the current state and combined with the prediction value, the optimal estimate value of the current state is:

[0061] X(k|k) = X(k|k - 1) + Kg(k)C(k) (b4)

[0062] To ensure that the robust Kalman filter can continuously filter the system, it is also necessary to update the covariance of X(k|k):

[0063] P(k|k) = (I - Kg(k)H)P(k|k - 1) (b5)

[0064] In equation (8): I is the identity matrix, and for the measurement mode of this system, I = 1; the robust Kalman filter is repeatedly executed on the data, and the finally obtained filtered data is denoted as g(t).

[0065] Furthermore, before the operation of the poultry dynamic weighing algorithm in step S3, an improved Robust Kalman filtering - Empirical Mode Decomposition (RKF - EMD) dynamic weighing algorithm is used for data filtering to remove outliers, including the following steps:

[0066] Assume that the collected original weight data is w(t), and w(t) is sorted to obtain the upper quartile Q3 and the lower quartile Q1; let the upper edge be H(t) and the lower edge be L(t), then:

[0067] IQR = Q3 - Q1 (c1)

[0068] H(t) = Q3 + k * IQR (c2)

[0069] L(t) = Q3 - k * IQR (c3) where k is the outlier factor, and k = 1.5 is taken to remove the abnormal data greater than the upper edge value or less than the lower edge value, and the cleaned data g(t) satisfies the following relationship:

[0070] L(t) < g(t) < H(t) (c4).

[0071] Compared with the prior art, the present invention has the following advantages:

[0072] Through reasonable scheme design and verification of the present invention, and comparison with traditional EMD decomposition, truncated mean method and wavelet transform algorithm, and through actual application monitoring in a poultry farm, it is found that the wireless dynamic adaptive weighing system suitable for poultry can well meet the requirements for weight detection in poultry farming while simultaneously possessing both speed and accuracy, and has high practicality. Data collected in broiler farms shows that the system can measure weight with a minimum error of less than 1% when the growth status of broilers in cages is relatively uniform, which is 1 / 3 of the error of manual weighing. Attached Figure Description

[0073] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0074] Figure 1 This is a schematic diagram of the field application of the weighing system of this invention.

[0075] Figure 2 This is a system hardware block diagram of Embodiment 1.

[0076] Figure 3 This is a schematic diagram of the weighing data entering the queue in Example 1.

[0077] Figure 4 This is a schematic diagram of the RKF-EMD dynamic weighing process in Example 1.

[0078] Figure 5 This is a comparison chart of the automatic peeling algorithm results in Example 1.

[0079] Figure 6 This is a line graph showing the process before and after outlier removal in Example 1.

[0080] Figure 7 This is a diagram showing the RKF-EMD experimental results of Example 1.

[0081] Figure 8 This is a system hardware block diagram of Embodiment 2.

[0082] Figure 9 This is a flowchart illustrating the principle of dynamic weighing process combining data cleaning and EMD in Example 2.

[0083] Figure 10 This is a comparison chart of the automatic peeling algorithm results in Example 2.

[0084] Figure 11This is the original data box plot of Example 2.

[0085] Figure 12 This is a box plot of Example 2 after outlier removal.

[0086] Figure 13 This is a graph showing the EMD experimental results of Example 2.

[0087] Figure 14 This is a 3D image of the actual weighing platform data collector.

[0088] Figure 15 This is a bottom view of the actual weighing platform data collector.

[0089] Figure 16 This is a picture of the circuit board of the weighing platform data collector.

[0090] Figure 17 This is a schematic diagram of the circuit board for the weighing platform data collector.

[0091] In the diagram: 1. Feeding cage; 2. Weighing platform data collector; 3. LoRa to RS485 circuit; 4. Smart gateway. Detailed Implementation

[0092] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this invention.

[0093] Example 1

[0094] like Figure 1 The diagram shows a field application of a wireless dynamic adaptive weighing system for poultry. The breeding environment is divided into a breeding house and an operating room. The smart gateway 4 and the LoRa to RS485 circuit 3 are placed in the operating room for easy access to data and modification of relevant breeding parameters by the breeders. In the breeding house, the weighing platform data collectors 2 are placed at fixed locations using a multi-point, spaced arrangement. Without any intervention, the weighing platform data collectors 2 transmit the collected data via their own LoRa to the LoRa to RS485 circuit 3, thus enabling data interaction between the data collectors 2 and the gateway. The smart gateway 4 and the data collectors adopt a star network topology. To prevent data conflicts during communication, a retransmission mechanism is incorporated into the system during data interaction to improve data transmission stability.

[0095] like Figure 2The diagram shows the hardware architecture of a wireless dynamic adaptive weighing system suitable for poultry. Considering the need for powerful data processing capabilities and an intuitive human-machine interface for such a system, the intelligent gateway 4 uses an STM32F429IGT6 as its core processor for algorithm calculations; the weighing platform data collector 2 uses an AT32F427F8P7 as its core, which has good AD performance and can minimize acquisition errors; and a 7-inch resistive LCD screen serves as the human-machine interface, allowing for clear and concise viewing of weighing data and setting of system parameters.

[0096] During the breeding process, manure and feed accumulate on the weighing platform over time, causing a shift in the baseline weight when the system is not in operation, thus increasing the result error. Therefore, this system innovatively proposes a queue-based automatic tare algorithm based on dynamic weighing, making the weighing algorithm more adaptive and robust. A diagram illustrating the entry of weighing data into the queue is shown below. Figure 3 As shown.

[0097] When the system starts, the initial weight value is recorded as z. The weight data collected by the system is then entered into a queue of length 10 according to a "first-out" rule. Each time data is enqueued, pairwise comparisons are performed. When the difference between any two pairs of data in the queue is less than 5g, the data in the queue is filtered to remove extreme values ​​and averaged to obtain the result w. The difference Δ is obtained by subtracting the average value from the initial value.

[0098] Δ=wz (1)

[0099] Where Δ is a signed number, a positive value represents poultry stepping onto the weighing platform, and a negative value represents poultry leaving the platform. At this point, the value of Δ and the initial weight value z are stored. Using w as the new initial value, the above steps are repeated. During this process, the following is obtained: n Each Δ and its corresponding initial weight z n Every so often, an actual poultry weight W is obtained through EMD decomposition. W is then compared with Δ. If the difference between the two is within 3g, the corresponding initial weight z is considered to be at that time. i This is the new initial state, which is used as the new baseline weight for weighing, thus achieving an automatic tare effect.

[0100] In poultry farming, overlapping and tiered cage systems are commonly used. Although the poultry's activity space is limited, their habits necessitate accurate weight measurement without affecting their normal activity. Simultaneously, to meet the data volume and accuracy requirements of the algorithm, a balance must be struck between program execution time and precision when designing this system. The MCU selected for this system possesses powerful computing capabilities; however, simply applying the EMD algorithm to the program would result in excessive computation and low accuracy. Therefore, filtering to remove outliers before algorithm execution and checking for accuracy requirements during execution are essential to exit algorithm calculations as early as possible and improve system efficiency.

[0101] This system employs the RKF-EMD dynamic weighing algorithm. Due to the characteristics of poultry and sensor noise, the obtained data is characterized by high noise and large oscillations, and the statistical characteristics of the noise are uncertain. Therefore, this invention uses a robust Kalman filter algorithm to filter and remove outliers. The flowchart of the RKF-EMD dynamic weighing algorithm is shown below. Figure 4 As shown.

[0102] For relatively stable dynamic measurement systems, Kalman filtering can be very effective. However, in the dynamic weighing process of poultry, due to the habits of poultry, large-amplitude noise, referred to as outliers, will be collected. If outliers are not removed, they will introduce significant deviations to the filtered estimates, affecting the accuracy of the final decomposition. Therefore, robustness is added to the traditional Kalman filter to remove outliers. Robust Kalman filtering consists of a prediction step and an update step.

[0103] Assuming k represents the current state of the system, and the process noise W(k) and system measurement noise V(k) are Gaussian noises with zero mean and covariances Q and R, respectively, the current state of the system can be obtained by Kalman filtering as follows:

[0104] X(k|k-1)=AX(k-1|k-1)+BU(k) (2)

[0105] Then, from equation (2), the covariance of X(k|k-1) can be obtained as:

[0106] P(k|k-1)=AP(k-1|k-1)A T +Q (3)

[0107] The prediction steps of Kalman filtering are implemented by equations (2) and (3) to obtain the state of the weighing system in the next time step.

[0108] In the update step, robustness is incorporated to remove outliers. First, the Kalman gain Kg(k) is obtained as follows:

[0109] Kg(k)=P(k|k-1)HT [HP(k|k - 1)H T +R] -1 (4)

[0110] The elimination of outliers is achieved through the observation residuals. The observation residuals represent the difference between the observed value and the predicted value, and are used to measure the deviation between the observed data and the predicted value. The larger this deviation, the greater the probability that the current data is an outlier. Denote the observation residual as C(k):

[0111] C(k) = Z(k) - HX(k|k - 1) (5)

[0112] At this time, a robust parameter is defined as m, which is used to control the sensitivity to outliers. The square root of the state covariance multiplied by the robust parameter is denoted as E(k), that is:

[0113]

[0114] Compare the absolute value of E(k) with the observation residual C(k):

[0115] a) When |C(k)| > E(k), it is considered that the current data is an outlier, and the state predicted value X(k|k - 1) is used to replace this data;

[0116] b) When |C(k)| < E(k), continue to execute the update step, and use the current data to correct the state estimate value.

[0117] When continuing to execute the update step, the prediction result of the current state can be obtained from the estimation of the system by equations (2) and (3). Combining the measurement value of the current state with the predicted value, the optimal estimated value of the current state is:

[0118] X(k|k) = X(k|k - 1) + Kg(k)C(k) (7)

[0119] To ensure that the robust Kalman filter can continuously filter the system, it is also necessary to update the covariance of X(k|k):

[0120] P(k|k) = (I - Kg(k)H)P(k|k - 1) (8)

[0121] In equation (8): I is the identity matrix, and for the measurement mode of this system, I = 1.

[0122] Repeatedly execute the robust Kalman filter on the data, and finally the filtered data is denoted as g(t).

[0123] To adapt to the habits of poultry, the weighing platform data collector continuously collects data while judging the poultry's physical characteristics. When consecutive similar data are detected, it is assumed that the poultry is in a stationary state. The median average filter is applied to this continuous data segment, and the average value is used as the final result. The robust Kalman filter and EMD algorithm are skipped to save system resources.

[0124] Because boundary effects exist during EMD decomposition, leading to errors in the final IMF results, it is necessary to address these boundary effects before decomposition to reduce systematic errors. The industry generally considers mirror extension to be the best method for resolving boundary effects. Mirror extension eliminates endpoint effects while also increasing the cardinality of the data, thus improving the accuracy of EMD decomposition.

[0125] When the birds are in a non-static state, the data after cleaning is g(t). Find the extreme points in g(t) from head to tail and from tail to head. Use the two extreme points as symmetrical points to perform mirror extension to obtain the data s(t).

[0126] By sorting and comparing, all maxima and minima in s(t) are found, and the upper envelope f is obtained through cubic sample interpolation. max (t) and lower envelope f min (t). Take the average of the upper and lower envelopes, denoted as m1(t), that is:

[0127]

[0128] Subtracting the original signal s(t) from m1(t) yields:

[0129] h1(t)=s(t)-m1(t) (10)

[0130] Determine whether h1(t) satisfies the following conditions:

[0131] a) The number of extreme points and zero-crossing points are equal or differ by at most 1;

[0132] b) The mean of the upper and lower envelopes is 0;

[0133] If h1(t) satisfies the above conditions, then h1(t) is the first IMF component, denoted as c1(t). Conversely, if h1(t) does not meet the above conditions, the above process is repeated using h1(t) as the new original signal, i.e.:

[0134] h2(t)=h1(t)-m2(t) (11)

[0135] Assuming that after repeating k times, the two conditions of the IMF are met, the first IMF component is obtained, namely:

[0136] c1(t)=h k (t) (12)

[0137] Removing c1(t) from the original signal yields the remaining signal:

[0138] r1(t)=s(t)-c1(t) (13)

[0139] After repeating the above process n times with r1(t) as the new original signal, s(t) is finally decomposed into n IMF components and a residual r representing the trend of the original signal. n (t):

[0140]

[0141] To prevent the EMD from continuously decomposing and causing the derived essential mode functions to lose their physical meaning of instantaneous frequency and amplitude, it is necessary to limit the number of signal iterations. This system uses the standard deviation criterion as the stopping criterion; iteration stops when the standard deviation is between 0.2 and 0.3. The formula is as follows:

[0142]

[0143] During EMD decomposition, spurious components sometimes appear—components that have no physical meaning or are unrelated to the signal—thus affecting the accuracy of the final weight result. This system uses the Pearson correlation coefficient to determine whether the current IMF (Individual Weight Default Factor) is spurious. If the similarity to the original data is below a threshold, it is considered a spurious IMF. The correlation coefficient function is as follows:

[0144]

[0145] This system needs to determine the correlation between each IMF component and the original signal to remove spurious components. Specifically, the correlation coefficient between each IMF component and the original signal is defined as follows:

[0146]

[0147] This system sets the threshold to one-tenth of the maximum value in the correlation coefficient sequence, that is:

[0148]

[0149] After removing spurious components using the correlation coefficient method, a one-dimensional array of residual quantities representing the signal trend is obtained. The average value of the residual quantities is calculated by median filtering, and the result is the true weight obtained from this dynamic weighing, denoted as W.

[0150] like Figures 14 to 17As shown, the weighing sensor used in the system's weighing platform data acquisition unit is model NA1-10kg, with an accuracy of 2.0mV / V±10%. The analog-to-digital converter chip is AD7190, with a maximum noise-free resolution of 22.5 bits and an output rate of 4.7Hz-4.8KHz. The high-precision and high-speed acquisition can ensure that the system's acquisition error is kept within 0.1%.

[0151] To verify the feasibility and accuracy of the proposed automatic tare algorithm, a comparative experiment was conducted using 500g and 1kg standard weights (M1 grade). The weights were read every hour inside the broiler farm, with each weight read 12 times to obtain the final results. The experimental results are as follows: Figure 5 As shown.

[0152] Experimental results show that the automatic tare algorithm proposed in this invention can maintain the baseline weight offset error within 1% without any external intervention. The results demonstrate that the proposed automatic tare algorithm has good performance, laying a reliable data foundation for subsequent data acquisition.

[0153] The data acquisition target of this invention is 22-day-old broilers. The system was placed inside the rearing cage and collected a total of 144 data packets within 15 seconds. The robust Kalman filter of this invention was used to remove outliers from these data packets. The resulting line graphs of the original data and those after outlier removal are shown below. Figure 6 As shown. (Through) Figure 6 The results show that the outliers in the collected data were effectively removed after being processed by robust Kalman filtering.

[0154] After robust Kalman filtering, the weight data was mirror-extended and then decomposed by EMD to obtain multiple IMFs and residuals. Finally, median filtering was applied to the residuals to obtain an average value of 902.7g. Under static conditions, the average weight of all broilers in the cage was measured to be 905.6g, with an overall error of 0.32%, meeting the system's accuracy design requirements. The RKF-EMD experimental results are as follows: Figure 7 As shown.

[0155] To highlight the advantages of this algorithm, weight data of infants aged 22-25 days were collected and compared with other dynamic weighing algorithms. The calculated weights and error comparison results are shown in Tables 1 and 2:

[0156] Table 1 Comparison of Algorithm Results

[0157]

[0158] Table 2 Comparison of Algorithm Result Errors

[0159]

[0160] The results in Tables 1 and 2 show that, under the same conditions, the RKF-EMD dynamic weighing algorithm presented in this paper maintains an error of less than 1% compared to the actual weight; the traditional EMD decomposition and truncated mean method maintain an error of 2%-5%; and the wavelet transform method, which decomposes and reconstructs the data to obtain the final average weight, maintains an error of 1%-2%. The dynamic weighing algorithms compared above all have larger errors than the algorithm presented in this paper. This is because poultry's habits of constantly moving up and down the weighing platform and moving around lead to many outliers in the data; at the same time, the rapid movement of poultry across the weighing platform also results in significant noise disturbances in the data, thus affecting the algorithm results. EMD is a method based on the local characteristics of the signal. These outliers introduce unnecessary noise and disturbances when constructing each mode function. When there are too many outliers, the disturbances and noise have a more significant impact on the results. Through comparative experiments, it was found that traditional EMD, truncated mean, and wavelet transform methods are not suitable for situations with very large data fluctuations. Excessive data fluctuations lead to increased errors in the final results, which limits their applicability to automatic weighing in poultry farming. In contrast, the algorithm in this system can preserve the local characteristics of the original data and reduce errors even when the poultry are constantly moving and generating undesirable data. This enables non-contact weighing of poultry, meeting the needs for weight monitoring in poultry farming.

[0161] Example 2

[0162] Based on Example 1, this example presents a wireless dynamic adaptive weighing verification method suitable for poultry farming. For example... Figure 8 The diagram shows a field application schematic of a wireless dynamic adaptive weighing system suitable for poultry. Weighing platform data collectors 2 are placed at multiple intervals to collect weight data, and the sampled data is used to statistically analyze the weight of poultry throughout the entire poultry shed. Without any intervention, the weighing platform data collectors 2 exchange the collected weight data with the gateway via a LoRa-to-RS485 circuit 3. Since the devices are in a star network topology, a retransmission mechanism is incorporated into the master-slave data exchange to prevent data conflicts during communication, thus improving the stability of data transmission.

[0163] like Figure 9 As shown, the hardware architecture of the wireless dynamic adaptive weighing system suitable for poultry is as follows: Considering that the wireless dynamic adaptive weighing system suitable for poultry requires powerful data computing and processing capabilities, as well as an intuitive human-machine interface, the intelligent gateway 4 of this system uses an STM32F429IGT6 as the core processor for algorithm calculation; the weighing platform data collector 2 uses an AT32F427F8P7 as its core, which has good AD performance and can minimize data acquisition errors; a 7-inch resistive LCD color screen serves as the human-machine interface, allowing for a simple and clear view of weighing data and setting of system parameters.

[0164] During the breeding process, manure and feed accumulate on the weighing platform over time, causing a shift in the baseline weight when the system is not in operation, increasing the result error. Therefore, this system innovatively proposes a queue-based automatic tare algorithm based on dynamic weighing, making the weighing algorithm more adaptive and robust. A diagram illustrating the entry of weighing data into the queue is shown below. Figure 3 As shown.

[0165] When the system starts, the initial weight value is recorded. The weight data collected by the system is then individually entered into a queue of length 10 according to a "first-in, first-out" (FIFO) rule. Each time data is added to the queue, pairwise comparisons are performed. When the difference between any two pairs of data in the queue is less than 5g, the data in the queue is filtered to remove extreme values ​​and averaged to obtain the result. The difference is obtained by subtracting the average value from the initial value.

[0166] Δ=wz

[0167] Here, Δ is a signed number, with a positive value representing poultry stepping onto the weighing platform and a negative value representing poultry leaving the platform. The Δ value and the initial weight z are stored at this point. Using w as the new initial value, the above steps are repeated. During this process, n Δ values ​​and their corresponding initial weights z are obtained. n Every so often, an actual poultry weight W is obtained through EMD decomposition. W is then compared with Δ. If the difference between the two is within 3g, the corresponding initial weight z is considered to be at that time. i This is the new initial state, which is used as the new baseline weight for weighing, thus achieving an automatic tare effect.

[0168] In poultry farming, overlapping and tiered cage systems are commonly used. Although the poultry's activity space is limited, their habits necessitate accurate weight measurement without affecting their normal activity. Simultaneously, to meet the data volume and accuracy requirements of the algorithm, a balance must be struck between program execution time and precision when designing this system. The MCU selected for this system possesses powerful computing capabilities; however, simply applying the EMD algorithm to the program would result in excessive computation and low accuracy. Therefore, data cleaning to remove outliers is essential before algorithm execution, and accuracy requirements must be checked during operation to exit algorithm calculations as early as possible, thereby improving system efficiency.

[0169] like Figure 10As shown, this system employs a dynamic weighing algorithm combining data cleaning and EMD. Data cleaning involves analyzing and processing large amounts of data to extract valuable information. Due to the habits of poultry and sensor noise, the obtained data is characterized by high noise and large oscillations. To preserve the original data characteristics, this system uses box plots for outlier detection.

[0170] According to the above algorithm, let the original weight data of a package be w(t). Sort w(t) to obtain the upper quartile Q3 and the lower quartile Q1. Let the upper edge be H(t) and the lower edge be L(t), then:

[0171] IQR = Q3 - Q1 (1)

[0172] H(t)=Q3+k*IQR (2)

[0173] L(t)=Q3-k*IQR (3)

[0174] Where k is the outlier factor, usually k = 1.5 is used to remove outlier data that is greater than the upper margin value or less than the lower margin value, resulting in cleaned data g(t), which satisfies the following relationship:

[0175] L(t) <g(t)<H(t) (4)

[0176] To adapt to the habits of poultry, the weighing platform data collector continuously collects data while judging the poultry's physical characteristics. When consecutive similar data are detected, it is assumed that the poultry is in a stationary state. The median average of this continuous data is then filtered, and the average value is used as the final result. Data cleaning and EMD algorithms are skipped to save system resources.

[0177] Because boundary effects exist during EMD decomposition, leading to errors in the final IMF results, it is necessary to address these boundary effects before decomposition to reduce systematic errors. The industry generally considers mirror extension to be the best method for resolving boundary effects. Mirror extension eliminates endpoint effects while also increasing the cardinality of the data, thus improving the accuracy of EMD decomposition.

[0178] When the birds are in a non-static state, the data after cleaning is g(t). Find the extreme points in g(t) from head to tail and from tail to head. Use the two extreme points as symmetrical points to perform mirror extension to obtain the data s(t).

[0179] By sorting and comparing, all maxima and minima in s(t) are found, and the upper envelope f is obtained through cubic sample interpolation. max (t) and lower envelope f min(t). Take the average of the upper and lower envelopes, denoted as m1(t), that is:

[0180]

[0181] Subtracting the original signal s(t) from m1(t) yields:

[0182] h1(t)=s(t)-m1(t) (6)

[0183] Determine whether h1(t) satisfies the following conditions:

[0184] ① The number of extreme points and zero-crossing points are equal or differ by at most 1;

[0185] ②The mean of the upper and lower envelopes is 0;

[0186] If h1(t) satisfies the above conditions, then h1(t) is the first IMF component, denoted as c1(t). Conversely, if h1(t) does not meet the above conditions, the above process is repeated using h1(t) as the new original signal, i.e.:

[0187] h2(t)=h1(t)-m2(t) (7)

[0188] Assuming that after repeating k times, the two conditions of the IMF are met, the first IMF component is obtained, namely:

[0189] c1(t)=h k (t) (8)

[0190] Removing c1(t) from the original signal yields the remaining signal:

[0191] r1(t)=s(t)-c1(t) (9)

[0192] After repeating the above process n times with r1(t) as the new original signal, s(t) is finally decomposed into n IMF components and a residual r representing the trend of the original signal. n (t):

[0193]

[0194] To prevent the EMD from continuously decomposing and causing the derived essential mode functions to lose their physical meaning of instantaneous frequency and amplitude, it is necessary to limit the number of signal iterations. This system uses the standard deviation criterion as the stopping criterion; iteration stops when the standard deviation is between 0.2 and 0.3. The formula is as follows:

[0195]

[0196] During EMD decomposition, spurious components sometimes appear—components that have no physical meaning or are unrelated to the signal—thus affecting the accuracy of the final weight result. This system uses the Pearson correlation coefficient to determine whether the current IMF (Individual Weight Default Factor) is spurious. If the similarity to the original data is below a threshold, it is considered a spurious IMF. The correlation coefficient function is as follows:

[0197]

[0198] This system needs to determine the correlation between each IMF component and the original signal to remove spurious components. Specifically, the correlation coefficient between each IMF component and the original signal is defined as follows:

[0199]

[0200] This system sets the threshold to one-tenth of the maximum value in the correlation coefficient sequence, that is:

[0201]

[0202] After removing spurious components using the correlation coefficient method, a one-dimensional array of residual quantities representing the signal trend is obtained. The average value of the residual quantities is calculated by median filtering, and the result is the true weight obtained from this dynamic weighing, denoted as W.

[0203] The weighing sensor used in the system's weighing platform is model NA1-20kg, with an accuracy of 2.0mV / V±10%. The analog-to-digital converter chip is AD7190, with a maximum noise-free resolution of 22.5 bits and an output rate of 4.7Hz-4.8KHz. The high-precision and high-speed acquisition can ensure that the system's acquisition error is kept within 0.1%.

[0204] To verify the feasibility and accuracy of the proposed automatic tare algorithm, a comparative experiment was conducted using 500g and 1kg standard weights (M2 grade). Weights were placed in the broiler house every hour to record the weights, with each weight recorded 12 times to obtain the final result. The experimental results are as follows: Figure 10 As shown.

[0205] Experimental results show that the automatic tare algorithm proposed in this invention can maintain the baseline weight offset error within 1% without any external intervention. The results demonstrate that the proposed automatic tare algorithm has good self-calibration performance, laying a reliable data foundation for subsequent data acquisition.

[0206] The data collection target of this invention is cage-raised broiler chickens aged 22 days. The system collected a total of 158 data packets within 15 seconds. The raw data box plot is shown below. Figure 11As shown in the figure. After box plot removal, it is clear that all outliers have been removed. The box plot after outlier removal is shown in the figure. Figure 12 As shown in the figure. After the weight data was removed, it was mirrored and then subjected to EMD decomposition to obtain multiple IMFs and residual amounts. Finally, the median value of the residual amounts was filtered to obtain an average value of 901.4g. The average weight of all broilers in the cage was measured to be 905.6g under static conditions, with an overall error of 0.46%, which meets the accuracy design requirements of the system. The EMD experimental results are as follows. Figure 13 As shown.

[0207] To highlight the advantages of this algorithm, weight data of infants aged 22-25 days were collected for comparison with other dynamic weighing algorithms. The calculated weights and error comparison results are shown in Tables 3 and 4:

[0208] Table 3 Comparison of Algorithm Results

[0209]

[0210] Table 4 Comparison of Algorithm Result Errors

[0211]

[0212] The results in Tables 3 and 4 show that, under the same conditions, the dynamic weighing algorithm combining data cleaning and improved EMD maintains an error of less than 1% compared to the actual weight; the traditional EMD decomposition and truncated mean method maintains an error of 2%-5%; and the wavelet transform method, which decomposes and reconstructs the data to obtain the final average weight, maintains an error of 1%-2%. The dynamic weighing algorithms compared above all have larger errors than the algorithm of this invention. This is because poultry's habits of constantly moving up and down the weighing platform and moving around lead to many outliers in the data; at the same time, the rapid movement of poultry across the weighing platform also results in significant noise disturbances in the data, thus affecting the algorithm results. EMD is a method based on the local characteristics of the signal. These outliers introduce unnecessary noise and disturbances when constructing each mode function, and when there are too many outliers, the disturbances and noise have a more significant impact on the results.

[0213] Through comparative experiments, it was found that traditional EMD, truncated mean, and wavelet transform methods are not suitable for situations with very large data fluctuations. Excessive data fluctuations lead to increased errors in the final results, which limits their applicability to automatic weighing in poultry farming. In contrast, the algorithm in this system can preserve the local characteristics of the original data and reduce errors even when the poultry are constantly moving and generating undesirable data. This enables non-contact weighing of poultry, meeting the needs for weight monitoring in poultry farming.

[0214] In summary, this invention introduces the operating principle, design scheme, and composition of a wireless dynamic adaptive weighing system suitable for poultry, as well as the principle, performance, and advantages of the improved algorithm. Through reasonable scheme design and verification of this invention, comparison with the traditional EMD decomposition algorithm, and practical application monitoring in a poultry farm, it is concluded that the wireless dynamic adaptive weighing system suitable for poultry can well meet the requirements for weight detection in poultry farming while simultaneously achieving both speed and accuracy, demonstrating high practicality.

[0215] Although the present invention has been described in detail with reference to the accompanying drawings and preferred embodiments, the invention is not limited thereto. Various equivalent modifications or substitutions can be made to the embodiments of the invention by those skilled in the art without departing from the spirit and essence of the invention, and such modifications or substitutions should all be within the scope of the invention. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the invention should also be covered within the protection scope of the invention. Therefore, the protection scope of the invention should be determined by the scope of the claims.

Claims

1. A wireless dynamic adaptive weighing verification method suitable for poultry farming, characterized in that, Includes the following steps: S1. Construction of the wireless dynamic adaptive weighing system: The system includes a feeding cage (1), which is set up in the breeding house in a multi-point interval manner. The weighing platform collector (2) is placed at a fixed point on the feeding cage (1). The weighing platform collector (2) has a built-in LoRa circuit. The LoRa circuit sends the collected weight data to the LoRa to RS485 circuit (3). Several weighing platform collectors (2) interact with the smart gateway (4) through a star network mode. The smart gateway (4) is used for system data calculation and processing. S2. The proposed automatic tare algorithm for weight data: Feces and feed continuously accumulate on the surface of the weighing platform collector (2). The weight data collected by the system is entered into the queue separately according to the first-out rule. Each time data is entered into the queue, the data in the queue are compared pairwise. When the difference between the pairs of data in the queue is within the range, the data in the queue is filtered by removing extreme values ​​and averaged to obtain the result. The average value is subtracted from the initial value to obtain the difference. A positive value indicates that poultry has been put on the weighing platform collector (2). Every once in a while, a real poultry weight is obtained by EMD decomposition and used as the new benchmark weight for weighing to achieve the automatic tare effect. S3. Preparations before running the poultry dynamic weighing algorithm: Before running the algorithm, data cleaning / filtering is performed to remove outliers and during operation, the accuracy requirements are checked to improve the system's operating efficiency. S4. Adaptive weighing of poultry dynamic weighing algorithm: The weighing platform collector (2) continuously collects data and judges the physical characteristics of poultry; when it detects continuous similar data, the poultry is in a static state, and the median average of this continuous data is filtered, and the average value is used as the final result, and the cleaning / filtering and EMD algorithm are skipped to save system resources; since the boundary effect exists during EMD decomposition, the final IMF result will have errors, so the boundary effect needs to be solved before decomposition to reduce system errors; by mirror extension, the endpoint effect is eliminated and the cardinality of the data is increased, which improves the accuracy of EMD decomposition, including the following steps: S41. When the birds are in a non-static state, the data after cleaning is g(t). Find the extreme points in g(t) from head to tail and from tail to head. Use the two extreme points as symmetrical points to perform mirror extension to obtain the data s(t). S42. Find all the maxima and minima in s(t) by sorting and comparing, and obtain the upper envelope f by cubic sample interpolation. max (t) and lower envelope f min (t); Take the average of the upper and lower envelopes, denoted as m1(t), that is: Subtracting the original signal s(t) from m1(t) yields: h1(t)=s(t)-m1(t) (2) S43. Determine whether h1(t) satisfies the following conditions: a) The number of extreme points and zero-crossing points are equal or differ by at most 1; b) The mean of the upper and lower envelopes is 0; If h1(t) satisfies the above conditions, then h1(t) is the first IMF component, denoted as c1(t); otherwise, h1(t) is used as a new original signal, and the above process is repeated, i.e.: h2(t)=h1(t)-m2(t) (3) S44. Assume that after repeating k times, the two conditions of the IMF are satisfied. At this point, the first IMF component is obtained, namely: c1(t)=h k (t) (4) Removing c1(t) from the original signal yields the remaining signal: r1(t)=s(t)-c1(t) (5) After repeating the above process n times with r1(t) as the new original signal, s(t) is finally decomposed into n IMF components and a residual r representing the trend of the original signal. n (t): S45. To prevent the EMD from continuously decomposing and causing the derived essential mode functions to lose their physical meaning of instantaneous frequency and instantaneous amplitude, it is necessary to limit the number of signal iterations. The standard deviation criterion is used as the stopping criterion; iteration stops when the standard deviation is between 0.2 and 0.

3. The formula is as follows: During the EMD decomposition process, spurious components may sometimes appear, which are components that have no physical meaning or are unrelated to the signal, thus affecting the accuracy of the final weight result. S5. The Pearson correlation coefficient is used to determine whether the current IMF is fraudulent. If the similarity to the original data is below a threshold, it is considered a fraudulent IMF. The correlation coefficient function is as follows: To remove spurious components, it is necessary to determine the correlation between each IMF component and the original signal. Specifically, the correlation coefficient between each IMF component and the original signal is defined as follows: Set the threshold to one-tenth of the maximum value in the correlation coefficient sequence, that is: After removing spurious components using the correlation coefficient method, a one-dimensional array of residual quantities representing the signal trend is obtained. The average value of the residual quantities is calculated by median filtering, and the result is the true weight obtained from this dynamic weighing, denoted as W.

2. The wireless dynamic adaptive weighing verification method for poultry farming as described in claim 1, characterized in that, In the wireless dynamic adaptive weighing system of step S1, the intelligent gateway (4) uses a processor of model STM32F429IGT6, and the weighing platform collector (2) uses a processor of model AT32F427F8P7. The intelligent gateway (4) integrates a human-machine interface, memory, Internet of Things circuit, RS484 circuit, USB circuit and LoRa to RS485 circuit (3); the weighing platform collector (2) integrates a DC-DC circuit, AD acquisition circuit, LoRa circuit, MOS circuit and pressure sensor. The intelligent gateway (4) is connected to at least one weighing platform collector (2) through the LoRa circuit.

3. The wireless dynamic adaptive weighing verification method for poultry farming as described in claim 1 or 2, characterized in that, In the wireless dynamic adaptive weighing system of step S1, the pressure sensor used by the weighing platform collector (2) is model NA1-20kg with an accuracy of 2.0mV / V±10%; the AD acquisition circuit uses AD7190 with a maximum noise-free resolution of 22.5 bits and an output rate of 4.7Hz-4.8KHz. The high-precision and high-speed acquisition ensures that the acquisition error of the system is kept within 0.1%.

4. The wireless dynamic adaptive weighing verification method for poultry farming as described in claim 1, characterized in that, The automatic tare algorithm for the weight data in step S2 adopts a queue-based automatic tare algorithm, namely: S21. When starting up, record the initial state weight value as z. The weight data collected by the system is separately entered into a queue with a data length of 10 according to the first-out rule. Each time data is enqueued, the data in the queue must be compared pairwise. S22. When the difference between any two data points in the queue is less than 5g, perform extreme value removal and averaging filtering on the data in the queue to obtain the result w. Subtract the average value w from the initial value to obtain the difference Δ. Δ=wz Δ is a signed number; a positive value represents an animal stepping onto the scale, and a negative value represents an animal leaving the scale. The Δ value and the initial weight z are stored. The above steps are repeated using w as the new initial value. In this process, n Δ values ​​and their corresponding initial weights z are obtained. n ; S23. Every so often, an actual animal weight W is obtained through EMD decomposition. W is compared with Δ. If the difference between the two is within 3g, the corresponding initial weight z is considered to be... i This is the new initial state, which is used as the new baseline weight for weighing to achieve an automatic tare effect.

5. The wireless dynamic adaptive weighing verification method for poultry farming as described in claim 4, characterized in that, Before the poultry dynamic weighing algorithm runs in step S3, an improved RKF-EMD dynamic weighing algorithm is used for data filtering to remove outliers, including the following steps: A. Prediction step: Assume that \(k\) represents the current state of the system. The process noise \(W(k)\) and the system measurement noise \(V(k)\) are Gaussian noises with zero mean and covariances \(Q\) and \(R\) respectively. The current state of the system can be obtained by Kalman filtering as follows: \(X(k|k - 1)=AX(k - 1|k - 1)+BU(k)\) (a1) Then, the covariance of \(X(k|k - 1)\) can be obtained from Equation (2) as: P(k|k-1)=AP(k-1|k-1)A T +Q (a2) The prediction step of Kalman filtering is realized by Equations (2) and (3) to obtain the state of the weighing system at the next time. B. Update step: First, the Kalman gain \(Kg(k)\) is obtained as: Kg(k)=P(k|k-1)H T [HP(k|k-1)H T +R] -1 (b1) The rejection of outliers is realized through the observation residual. The observation residual represents the difference between the observed value and the predicted value, and is used to measure the deviation between the observed data and the predicted value. The larger the deviation, the greater the probability that the current data is an outlier. Denote the observation residual as \(C(k)\): \(C(k)=Z(k)-HX(k|k - 1)\) (b2) At this time, a robust parameter \(m\) is defined to control the sensitivity to outliers. The product of the robust parameter and the square root of the state covariance is denoted as \(E(k)\), that is: Compare the absolute value of \(E(k)\) with the observation residual \(C(k)\): a) When \(|C(k)|>E(k)\), it is considered that the current data is an outlier, and the state predicted value \(X(k|k - 1)\) is used to replace this data. b) When \(|C(k)|<E(k)\), continue to execute the update step, and use the current data to correct the state estimate value. When continuing to execute the update step B, the predicted result of the current state is obtained by the estimation of Equations (2) and (3). According to the measured value of the current state and combined with the predicted value, the optimal estimated value of the current state is: \(X(k|k)=X(k|k - 1)+Kg(k)C(k)\) (b4) To ensure that the robust Kalman filter can continuously filter the system, it is also necessary to update the covariance of \(X(k|k)\): \(P(k|k)=(I - Kg(k)H)P(k|k - 1)\) (b5) In Equation (8): \(I\) is the identity matrix, and for the measurement mode of this system \(I = 1\).

6. The wireless dynamic adaptive weighing verification method for poultry farming as described in claim 4, characterized in that, Repeat the robust Kalman filter for the data, and finally the filtered data is denoted as \(g(t)\). Before the operation of the poultry dynamic weighing algorithm in step S3, a dynamic weighing algorithm combining data cleaning and EMD is used for data cleaning to remove outliers, including the following steps: Assume that the original weight data collected is \(w(t)\). Sort \(w(t)\) to obtain the upper quartile \(Q3\) and the lower quartile \(Q1\). Let the upper edge be \(H(t)\) and the lower edge be \(L(t)\), then: \(IQR = Q3 - Q1\) (c1) \(H(t)=Q_{3}+k*IQR\) (c2) \(L(t)=Q_{1}-k*IQR\) (c3) Among them, \(k\) is the outlier factor, and \(k = 1.5\) is taken to remove the outlier data greater than the upper edge value or less than the lower edge value, and the cleaned data \(g(t)\) is obtained. \(g(t)\) satisfies the following relationship: \(L(t)<g(t)<H(t)\) (c4).

Images