Poultry house multi-target sensor stationing optimization method based on conditional constraints

Abstract

The invention discloses a poultry house multi-target sensor stationing optimization method based on conditional constraints, and belongs to the technical field of poultry house environment monitoring. K types of environment dynamic sensors are placed in a poultry house according to N initial sites where environment information needs to be obtained, the optimal arrangement method of the sensors is researched under the combined action of the types and positions of the multiple sensors, robot constraint and cost constraint, errors, information gains, coverage rates and penalty term factors are integrated, and the optimal arrangement of the sensors is achieved. According to the method, the arrangement positions of the monitoring points can be determined according to self-defined evaluation preferences, key environment monitoring points of the poultry house are found, the purpose of reducing the number of the environment monitoring points of the poultry house is expected to be achieved, and the efficiency of poultry house environment monitoring is improved.

Description

Technical Field

[0001] The invention relates to the technical field of poultry house environment monitoring, and in particular to a poultry house multi-target sensor layout optimization method based on conditional constraints. Background Art

[0002] In the process of intensive poultry farming, it is necessary to monitor important environmental indicators of poultry houses. The precise environmental control system is an important part of maintaining a good breeding environment in the house. It mainly collects environmental information through sensors distributed in the poultry house, and then controls the corresponding environmental control equipment through environmental control strategies to keep the sensor site area environment in the house at the set threshold. The sensor point layout scheme currently used is mainly based on the plum blossom method and the diagonal method, and the average value is mainly used as the control indicator, but the dynamic changes of environmental factors in the poultry house with the influence of external factors are ignored. In particular, with the continuous deepening of the demand for refined breeding, environmental control has gradually moved towards the study of precise control of microclimate environment. Therefore, the layout of sensors should be representative and able to reflect the actual environmental state of the microclimate environment in the poultry house to the greatest extent at a certain cost. In addition, with the development of artificial intelligence and informatization, although inspection robots have played an important role in the process of poultry house breeding, their disadvantages are limited time and space monitoring problems, and they cannot monitor the environment of the whole house at the same time. In addition, farms also have different positioning requirements for inspection robots. Therefore, it is of great significance to combine the advantages of environmental sensors (static) and inspection robots (dynamic) to improve the efficiency and effect of whole-house monitoring. At present, many studies focus on the placement of fixed sensor monitoring points, without considering the constraints and synergies of multiple sensors, costs, and other collaborative devices. Summary of the invention

[0003] The purpose of the present invention is to provide a poultry house multi-objective sensor layout optimization method based on conditional constraints, and to study the sensor layout method under the joint effects of multi-sensor types and positions, robot constraints, and cost constraints. This method can determine the layout positions of monitoring points based on user-defined evaluation preferences, and find key environmental monitoring points in the poultry house, in order to achieve the purpose of reducing the number of poultry house environmental monitoring points, improve the efficiency of poultry house environmental monitoring, and solve the shortcomings of the prior art.

[0004] To achieve the above object, the present invention provides the following technical solutions:

[0005] A method for optimizing the placement of multi-objective sensors in a poultry house based on conditional constraints comprises the following steps:

[0006] S1: Place K types of environmental sensors at N initial locations in the poultry house where environmental information is to be obtained, and collect prior information on the target structure, including: obtaining a sensor data matrix D(t) of dimension N*K at time point t, where each element d i，k (t) represents the data value of the k-th sensor at the i-th monitoring point at time t;

[0007] S2: Collect the operating status information of the external environment and environmental control equipment as the characteristics for predicting the environmental factors in the poultry house:

[0008]

[0009] Where T out (t), T out (t) and V out (t) is the temperature, humidity and wind speed data outside the house, W angle (t), F status (t) and M status (t) is the operating status of the poultry house fan, wet curtain and heater;

[0010] S3: For each sensor data point i and characteristic factor D in the sensor data matrix D(t) ext (t) are used as inputs to predict the data of other sensor sites of the same type through the machine learning model, and then the error is calculated to obtain the error matrix E(t) with a dimension of N*N*K;

[0011] S4: The environmental monitoring capability of the inspection robot in the poultry house is coordinated with the fixed sensor in the poultry house to collect the coverage area C(t) of the inspection robot at each time point t of the fixed sensor:

[0012]

[0013] S5: Determine the number of fixed sensors in the area actually covered by the robot at time point t by checking the coordinates of the patrol robot Calculate the coverage rate R of the environmental dynamic sensors carried by the inspection robot in one day cover =n*f / 1440, where n is the duration of each inspection by the inspection robot, and f is the inspection frequency of the inspection robot's constraint condition;

[0014] S6: The coverage rate R of the inspection robot in one day cover and the number of environmental dynamic sensors covered in time t N cover (t), the number of replaceable fixed sensors is R cover *N cover (t);

[0015] S7: Calculate the information gain IG for each sensor position i，k =H(X rest )-H(X rest |S i，k ),in is a measure of the initial uncertainty of the system, X rest is the set of possible data values ​​of the remaining monitoring points, P(x rest ) is the probability of a certain data value of the remaining points, IG i，k is the initial entropy of the system H(X rest ) and the remaining entropy H(X rest |S i，k ) difference;

[0016] S8: Calculate the comprehensive evaluation index for each type of sensor Where k represents the sensor type, ε k represents the average prediction error of each type of sensor, IG k is the average information gain of various sensors W k is the importance weight of each type of sensor, α is the adjustment parameter of prediction error and information gain;

[0017] S9: Calculate the global priority, that is, under the budget constraint, allocate the quantity of each type of environmental dynamic sensors and set the objective function of the global priority where n k is the number of each sensor, and the constraint is And n k ≤NR cover *N cover (t), where C k The unit cost of each type of sensor is solved by mixed integer linear programming. k ;

[0018] S10: Perform local optimization, i.e., the selection of specific locations of each type of sensor: set an initial error threshold ε0, which represents the error of the desired selected sensor when predicting other points;

[0019] S11: Based on the error matrix, find the sensor with the smallest prediction error and the highest coverage for other sites;

[0020] S12: Combine the standardized error matrix and information gain to construct a comprehensive score for each sensor type and calculate the comprehensive evaluation index for each location Where β is a weight used to adjust the importance of information gain and error;

[0021] S13: After selecting the first sensor, mark all points covered by the sensor as covered, remove the already coverable points, recalculate the error matrix and information gain of the remaining points, repeat steps S11 to S12, and so on to sort all monitoring points of each sensor type k;

[0022] S14: For monitoring points with large prediction errors, add a penalty term P(ε i，k ), P(ε i，k ) is based on the maximum value of the prediction error of each sensor;

[0023] S15: Constructing the objective function Choose the best placement for each type of sensor.

[0024] Furthermore, each element e in the error matrix E(t) in S3 i，k (t) represents the prediction error of the k-th sensor at the i-th monitoring point at time t, e i，k (t)=|d i，k (t)-P i，k (t)|, where d i，k (t) is the actual observed value, and Pi,k(t) is the predicted value.

[0025] Furthermore, the coverage R in S5 cover Close to 1, it means that the inspection robot’s detection time is long enough and the coverage rate R cover If it is close to 0, it means that the inspection robot is not working most of the time and its data contribution is limited.

[0026] Furthermore, the information gain IG at all time points in S7 i，k This can reflect the improved prediction accuracy achieved by increasing the number of fixed sensor locations.

[0027] Furthermore, the specific method in S11 is: calculate the error coverage of each point for the temperature, humidity, wind speed and carbon dioxide sensors respectively. s [i, k], that is, the error of predicting other points is less than ε0:

[0028]

[0029] Where 1(ε i，k ≤ε0) means that if the prediction error is less than the threshold, the result is 1, otherwise it is 0. The selection of a certain type of sensor means the selection of coverage s [i, k] The i with the largest value has priority as the first sensor location of this type.

[0030] Furthermore, the objective function constructed in S15 combines error, information gain, coverage and penalty factors. The constraints of the objective function are: nk is the total number of sensors of this type.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] The multi-objective sensor layout optimization method for a poultry house based on conditional constraints of the present invention studies the sensor layout method under the joint effects of multiple sensor types and positions, robot constraints, and cost constraints. Through this method, the layout positions of monitoring points can be determined according to user-defined evaluation preferences, and key environmental monitoring points of the poultry house can be found, so as to achieve the purpose of reducing the number of environmental monitoring points in the poultry house and improving the efficiency of environmental monitoring in the poultry house. DETAILED DESCRIPTION

[0033] The embodiments of the present invention will be described in detail below, however, the embodiments of the present invention are not limited thereto. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0034] An embodiment of the present invention provides a method for optimizing the placement of multi-target sensors in a poultry house based on conditional constraints, comprising the following steps:

[0035] S1: Place K types of environmental dynamic sensors at N initial locations in the poultry house where environmental information is to be obtained, collect prior information on the target structure, and obtain a data matrix D(t) of all sensors with a dimension of N*K at time point t, where each element d i，k (t) represents the data value of the kth sensor at the i-th monitoring point at time t. For example, the sensor data matrix at time t is:

[0036]

[0037] Where D T (t), D H (t), D H (t) and D C (t) represents the data of the temperature, humidity, wind speed and carbon dioxide sensors in the house at time t.

[0038] S2: Since the external environment and the operating status of the environmental control equipment will also affect the environmental distribution in the house, this part of information is collected at the same time as the characteristics of predicting the environmental factors in the house:

[0039]

[0040] Where T out (t), Tout (t) and V out (t) is the temperature, humidity and wind speed data outside the house, W angle (t), F status (t) and M status (t) is the operating status of the poultry house fan, wet curtain and heater;

[0041] S3: For each sensor data point i and D(t) ext (t) are used as inputs, and the data of other sensor sites of the same type are predicted through the machine learning model. Then, the error is calculated to obtain the error matrix E(t) with a dimension of N*N*K. i，k (t) represents the prediction error e of the k-th sensor at the i-th monitoring point at time t. i，k (t)=|d i，k (t)-P i，k (t)|, where d i，k (t) is the actual observed value, P i，k (t) is the predicted value. For example, for D at time point t T (t) Temperature field matrix. Assuming that the initial position is 60 points, each point i (i = 1, 2, ..., 60) and D ext (t) is used as input to predict the temperature values ​​of other points, and then the prediction error of each point is calculated, and finally a 60*60 error matrix E is obtained;

[0042] S4: The purpose of using patrol robots in different chicken houses is different, so it cannot be used as a quantitative measure, but its environmental monitoring ability can be coordinated with static sensors (fixed). The constraints of the patrol robot are that the patrol frequency is f and the duration of each patrol is n, in minutes. Collect the coverage rate of the site of the patrol robot during the monitoring interval of the fixed sensor. Assume that the coverage area C(t) of the collection robot of the fixed sensor at each time point t is:

[0043]

[0044] S5: Since the data monitored by the inspection robot at time point t is not necessarily the position where the static sensors completely overlap, there will be a certain deviation. Therefore, the number of fixed environmental dynamic sensors in the actual coverage area of ​​the robot at time point t is first determined by the robot's coordinates. Calculate the duration coverage rate R of the robot's environmental dynamic sensor (dynamic) in one day cover =n*f / 1440. If the coverage rate is close to 1, it means that the inspection robot has a long enough detection time. If the coverage rate is close to 0, it means that the robot is not working most of the time and the data contribution is limited.

[0045] S6: The coverage rate R of the robot within one day cover , and the number of environmental dynamic sensors covered in time t N cover (t), so the number of fixed sensors that can be replaced is R cover *N cover (t);

[0046] S7: Calculate the information gain IG for each sensor position i，k =H(X rest )-H(X rest |S i，k ),in is a measure of the initial uncertainty of the system, X rest is the set of possible data values ​​of the remaining monitoring points, P(x rest ) is the probability IG of a certain data value of the remaining points i，k , is the initial entropy of the system H(X rest ) and the remaining entropy H(X rest |S i，k ) The information gain IG of all time points is considered comprehensively. i，k It can reflect the improved prediction accuracy by increasing the fixed sensor locations;

[0047] S8: Since there are many types of environmental dynamic sensors, they have different focuses according to their monitoring needs. Therefore, the global priority is calculated first, that is, the number of each sensor is allocated under the budget constraint. The comprehensive evaluation of each sensor is calculated. Where k represents the sensor type, ε k represents the average prediction error of each type of sensor, IG k is the average information gain of various sensors w k is the importance weight of each type of sensor, α is the adjustment parameter of prediction error and information gain;

[0048] S9: Objective function for setting global priority where n k is the number of each sensor. The constraints are And n k ≤NR cover *N cover (t), where C k For the unit cost of each type of sensor, solve n by mixed integer linear programming k ;

[0049] S10: Then perform local optimization, i.e., select the specific location of each type of sensor. Set an initial error threshold ε0, which represents the error of the desired selected sensor when predicting other points;

[0050] S11: Based on the error matrix, find the sensor with the smallest prediction error and the highest coverage rate for other sites; specifically, calculate the error coverage rate of each point for the temperature, humidity, wind speed and carbon dioxide sensors, that is, predict that the error of other points is less than ε0:

[0051]

[0052] Where 1(ε i，k ≤ε0) means that if the prediction error is less than the threshold, the result is 1, otherwise it is 0. The selection of a certain type of sensor means the selection of coverage s [i, k] The largest i has priority over the first sensor location of this type;

[0053] S12: Combine the standardized error matrix and information gain to construct a comprehensive score for each sensor type. Calculate the comprehensive evaluation for each location Where β is a weight used to adjust the importance of information gain and error;

[0054] S13: After selecting the first sensor, mark all points covered by the sensor as covered, and remove the already coverable points. Recalculate the error matrix and information gain of the remaining points, repeat steps S11 to S12, and so on to sort all monitoring points of each sensor type k;

[0055] S14: In order not to ignore those monitoring points with large prediction errors, a penalty term P(ε i，k The penalty term is based on the maximum value of the prediction error of each sensor, ensuring that the points with large errors can be focused on;

[0056] S15: Constructing the objective function The optimal layout position of each type of sensor is selected, which combines the error, information gain, coverage and penalty factors. The constraint condition of this objective function is n k is the total number of sensors of this type.

[0057] To sum up: the embodiment of the present invention provides a poultry house multi-objective sensor layout optimization method based on conditional constraints, which studies the sensor layout method under the joint effects of multiple sensor types and positions, robot constraints, and cost constraints. Through this method, the monitoring point layout positions can be determined according to customized evaluation preferences, and the key environmental monitoring points of the poultry house can be found, so as to achieve the purpose of reducing the number of poultry house environmental monitoring points and improving the efficiency of poultry house environmental monitoring.

[0058] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for optimizing the placement of multi-target sensors in poultry houses based on conditional constraints, characterized in that: The following steps are involved: S1: Place K types of environmental sensors at N initial locations in the poultry house where environmental information is to be obtained, and collect prior information on the target structure, including: obtaining a sensor data matrix D(t) of dimension N*K at time point t, where each element d i，k (t) represents the data value of the k-th sensor at the i-th monitoring point at time t; S2: Collect the operating status information of the external environment and environmental control equipment as the characteristics for predicting the environmental factors in the poultry house: Where T out (t), T out (t) and V out (t) is the temperature, humidity and wind speed data outside the house, W angle (t), F status (t) and M status (t) is the operating status of the poultry house fan, wet curtain and heater; S3: For each sensor data point i and characteristic factor D in the sensor data matrix D(t) ext (t) are used as inputs to predict the data of other sensor sites of the same type through the machine learning model, and then the error is calculated to obtain the error matrix E(t) with a dimension of N*N*K; S4: The environmental monitoring capability of the inspection robot in the poultry house is coordinated with the fixed sensors in the poultry house to collect the coverage area C(t) of the inspection robot at each time point t of the fixed sensors: S5: Determine the number of fixed sensors in the area actually covered by the robot at time point t by checking the coordinates of the robot Calculate the coverage rate R of the environmental dynamic sensors carried by the inspection robot in a day cover =n*f / 1440, where n is the duration of each inspection by the inspection robot, and f is the inspection frequency of the inspection robot's constraint condition; S6: The coverage rate R of the inspection robot in one day cover and the number of environmental dynamic sensors covered in time t N cover (t), the number of replaceable fixed sensors is R cover *N cover (t); S7: Calculate the information gain IG for each sensor position i，k =H(X rest )-H(X rest |S i，k ),in is a measure of the initial uncertainty of the system, X rest is the set of possible data values ​​of the remaining monitoring points, P(x rest ) is the probability of a certain data value of the remaining points, IG i，k is the initial entropy of the system H(X rest ) and the remaining entropy H(X rest |S i，k ) difference; S8: Calculate the comprehensive evaluation index for each type of sensor Where k represents the type of sensor, ε k represents the average prediction error of each type of sensor, IG k is the average information gain of various sensors W k is the importance weight of each type of sensor, α is the adjustment parameter of prediction error and information gain; S9: Calculate the global priority, that is, under the budget constraint, allocate the quantity of each type of environmental dynamic sensors and set the objective function of the global priority where n k is the number of each sensor, and the constraint is And n k ≤NR cover *N cover (t), where C k For the unit cost of each type of sensor, solve n by mixed integer linear programming k ; S10: Perform local optimization, i.e., the selection of specific locations of each type of sensor: set an initial error threshold ε0, which represents the error of the desired selected sensor when predicting other points; S11: Based on the error matrix, find the sensor with the smallest prediction error and the highest coverage for other sites; S12: Combine the standardized error matrix and information gain to construct a comprehensive score for each sensor type and calculate the comprehensive evaluation index for each location Where β is a weight used to adjust the importance of information gain and error; S13: After selecting the first sensor, mark all points covered by the sensor as covered, remove the already coverable points, recalculate the error matrix and information gain of the remaining points, repeat steps S11 to S12, and so on to sort all monitoring points of each sensor type k; S14: For monitoring points with large prediction errors, add a penalty term P(ε i，k ), P(ε i，k ) is based on the maximum value of the prediction error of each sensor; S15: Constructing the objective function Select the best placement for each type of sensor.

2. The method for optimizing the placement of multi-target sensors in a poultry house based on conditional constraints according to claim 1, characterized in that: Each element e in the error matrix E(t) in S3 i，k (t) represents the prediction error of the k-th sensor at the i-th monitoring point at time t, e i，k (t)=|d i，k (t)-P i，k (t)|, where d i，k (t) is the actual observed value, P i，k (t) is the predicted value.

3. The method for optimizing the placement of multi-target sensors in a poultry house based on conditional constraints according to claim 1, characterized in that: S5 medium coverage R cover Close to 1, it means that the inspection robot’s detection time is long enough and the coverage rate R cover If it is close to 0, it means that the inspection robot is not working most of the time and its data contribution is limited.

4. The method for optimizing the placement of multi-target sensors in a poultry house based on conditional constraints according to claim 1, characterized in that: Information gain IG at all time points in S7 i，k This can reflect the improved prediction accuracy achieved by increasing the number of fixed sensor locations.

5. The method for optimizing the placement of multi-target sensors in a poultry house based on conditional constraints according to claim 1, characterized in that: The specific method in S11 is: calculate the error coverage of each point for temperature, humidity, wind speed and carbon dioxide sensors respectively s [i, k], that is, the error of predicting other points is less than ε0: Where 1(ε i，k ≤ε0) means that if the prediction error is less than the threshold, the result is 1, otherwise it is 0. The selection of a certain type of sensor means the selection of coverage s [i, k] The i with the largest value has priority as the first sensor location of this type.

6. The method for optimizing the placement of multi-target sensors in a poultry house based on conditional constraints according to claim 1, characterized in that: The objective function constructed in S15 combines error, information gain, coverage and penalty factors. The constraints of this objective function are: n k is the total number of sensors of this type.

Images