CN111726407B - A fog computing monitoring technology for the cultivation of famous flowers and medicinal plants in an intelligent plant factory environment - Google Patents

Abstract

The invention relates to a fog computing monitoring technology for the cultivation of famous flowers and medicinal plants under the environment of an intelligent plant factory. Module; Step 2. After each local device collects the sensor data, it estimates the subsequent data scale locally through a lightweight data scale estimation model to adjust the forwarding amount, and at the same time uses the local adaptive sampling model and adaptive filtering model. The sensor data is processed, and finally the processed sensor data is forwarded to the cloud server through the forwarding device according to the adjusted forwarding amount. The beneficial effects of the present invention are: the present invention estimates the transmission flow based on the lightweight estimation model for the environmental parameter data collected by the sensor during the growth process of the famous flowers, filters the sensing data flow through adaptive sampling and adaptive filtering technology, and uses fog The calculation enables real-time monitoring of the appropriate amount of effective sensor data.

Description

A fog computing monitoring technology for the cultivation of famous flowers and medicinal plants in an intelligent plant factory environment

technical field

The invention relates to the field of monitoring technology based on fog computing mode, and particularly includes a fog computing monitoring technology for cultivation of famous flowers and medicinal plants in an environment of an intelligent plant factory; an approximate estimation of low power consumption by establishing sensing data collection in the environment of a plant factory Model, adaptive sampling model and adaptive filtering model to realize the monitoring of environmental parameters for the cultivation of famous flowers and medicinal plants in the environment of intelligent plant factories.

Background technique

With the development of modern biotechnology, planting technology and Internet of Things technology, plant factories use thermal insulation and opaque materials as the enclosure structure, strictly control the gas exchange with the external environment, and use artificial light sources to provide light for plants, making the system The internal energy and material flow form a self-circulating regeneration system with the air circulation, which not only helps to increase the indoor carbon dioxide concentration, promotes photosynthesis to shorten the plant cultivation cycle, but also prevents bacteria and insect pests from entering the room, making the product cleaner. Compared with the traditional artificial cultivation method, the fine cultivation of plant factories is not affected by the outside world and can be precisely controlled. It can carry out high-density plant cultivation and realize the annual and stable production of high-quality plants. In addition, the location of plant factories is less restricted, and production can be carried out in cities according to local conditions, reducing environmental pollution and storage problems caused by automobile transportation in the logistics link.

The key to the factory-based fine cultivation of famous flowers and medicinal plants is to provide the special environment required for the growth of different plant varieties at different growth stages, and to accelerate the growth of plants to shorten the cultivation period. Therefore, monitoring the information of various environmental objects in plant factories (such as temperature, light, water, and soil pH, etc.) is the premise and core of the factory-based fine cultivation of famous flowers and medicinal plants. The existing monitoring technology for the cultivation of famous flowers and medicinal plants in fine chemical plants can be roughly divided into two steps: 1) Various environmental sensors collect relevant information according to preset frequencies, such as temperature, light, moisture and soil. pH, etc.; 2) The collected information is transmitted to the background monitoring center by wireless transmission methods such as 4G, WiFi or Zigbee. However, the amount of data generated by various environmental sensors in plant factories has increased sharply every day, and plant fine chemical factory cultivation has put forward higher requirements for the response time and security of environmental information monitoring technology. Although the existing monitoring technology for fine cultivation in plant factories provides an efficient computing platform for various environmental perception data processing, the current growth rate of network bandwidth is far from keeping up with the growth rate of data, and the decline rate of network bandwidth cost is faster than that of CPU. The cost of hardware resources such as memory and memory declines at a much slower rate, and at the same time, the complex network environment makes it difficult to achieve a breakthrough in network latency. Therefore, the existing monitoring technology for the cultivation of famous flowers and medicinal plants in fine chemical plants needs to solve the two bottlenecks of network bandwidth and delay.

SUMMARY OF THE INVENTION

The purpose of the present invention is to overcome the deficiencies in the prior art, and to provide a fog computing monitoring technology for the cultivation of famous flowers and medicinal plants in an intelligent plant factory environment.

The monitoring system for the cultivation of famous flowers and medicinal plants in the intelligent plant factory environment includes: a plant factory data acquisition module, a fog acquisition module, a cloud storage module and a monitoring system visualization module; the plant factory data acquisition module is connected through WiFi or Bluetooth. The fog collecting module is connected for data transmission, the fog collecting module uploads data to the cloud storage module, and the monitoring system visualization module reads data from the cloud storage module.

Preferably, the plant factory data collection module is provided with sensors, the sensors include a temperature sensor, a carbon dioxide sensor and a humidity sensor; the fog collection module is used for data scale estimation, adaptive sampling and adaptive filtering; the The cloud storage module is used for uploading data to the cloud and backing up the data locally; the monitoring system visualization module is used for environmental data visualization, plant growth visualization and nutrient solution status visualization.

The fog computing monitoring technology of the monitoring system of the famous flower and medicinal plant cultivation factory in the intelligent plant factory environment includes the following steps:

Step 1. Build a sensor network, divide the plant factory into areas, and divide the monitoring area for each local fog computing device. The plant factory data acquisition module deploys several local devices in each area to collect sensing data such as temperature, humidity, and carbon dioxide. ; Upload the collected sensor data to the fog collection module, such as temperature and humidity, carbon dioxide and soil pH, etc.; for subsequent forwarding;

Step 2. After each local device collects the sensor data, it estimates the subsequent data scale locally through the lightweight data scale estimation model to adjust the forwarding amount, and at the same time uses the local adaptive sampling model and adaptive filtering model to adjust the forwarding amount. The sensor data is processed, and finally the processed sensor data is forwarded to the cloud server through the forwarding device according to the adjusted forwarding amount;

Step 2.1. According to the scale of the sensing data, estimate the subsequent data scale through the probability exponentially weighted moving average model and trend detection, update the probability exponentially weighted moving average model online and adjust the amount of data forwarding:

_{Deploy a} lightweight data scale estimation model on the local device, and use the monitoring data of a specific environmental monitoring device in the past period to predict its future monitoring data; The distance δ _i is defined as follows:

δ _i =|v _i -v _i-1 |

Update the change of the local reference running time of the sensing data stream ρ(M) with the distance δ _i , calculate the current change of the sensing data stream by moving average, denoted as μ _i , and denote the distance of the next two data points as δ _i+1 ; the distance between two consecutive values represents the change of the sensory data flow; the approximate prediction of the monitoring sensory data is carried out using the probability exponentially weighted moving average model, and the weighting factor (0<α<1) is introduced to change exponentially The mode downweights older sensory data as follows:

In the above formula, u ₁ is initialized with δ ₁ and updated iteratively later; although the exponentially weighted moving average model is more suitable for the monitoring needs of actual plant factories, its response to transient changes is unstable, so it cannot always be assumed Only exponential weighting exists. Specifically, if a probabilistic exponentially weighted moving average model encounters a sudden peak after a longer period of stabilization, and this abrupt peak is followed by a stabilization period, the probabilistic exponentially weighted moving average model retains the peak ; this will lead to overestimation of the subsequent δ _i , which will affect the accuracy of the approximate data size estimation model;

Step 2.2. Deploy the adaptive sampling model for fine cultivation monitoring of plant factories on the local device, and calculate the confidence level according to the scale of the sensor data. After comparing with the user-defined inaccuracy, adjust the sampling period and sample the sensor data. ; The core of adaptive sampling is the process of dynamically adjusting the sampling periodicity T _i based on changes in the monitoring sensor data flow of the plant factory, while the monitoring accuracy still meets the accuracy requirements given by the user;

Furthermore, compared to the stepwise technique that adjusts the sampling rate only based on a step function (eg, T _i+1 ←T _i ±T _step , where T _step is the confidence interval radius); the proposed adaptive algorithm can The metric flow of the plant responds efficiently within the appropriate range [T _min , T _max ]; step 2.3, deploy the adaptive filtering model for fine cultivation monitoring of plant factories on the local device, and filter the sampled sensor data: The filtering range is expressed as R, if the value v _i ∈ [v _i-1 -R,v _i-1 +R], the current data point d _i of v _i is filtered; and the monitoring sensor data flow changes dynamically according to the Fano factor Adjust the filtering range R, and adjust the sampling period according to the filtering results;

Step 2.4, deploying an adaptive forwarding model for fine cultivation monitoring of plant factories on the local device, forwarding the sampled or filtered data according to the forwarding amount estimated by the data scale estimation model, and transmitting the data to the cloud server;

Step 3. The cloud server receives and parses the sensory data forwarded by the local device, saves different types of sensory data to the local database, and displays the plant growth status data or plant growth environment data on the device through a visual chart, to the management personnel. Provides real-time data monitoring.

Preferably, in the step 1, the local device includes a sensor.

Preferably, in step 2.1, in order to solve the problem that the probability exponentially weighted moving average model encounters a sudden data peak change, the probability exponentially weighted moving average model is solved by two steps of probability weighting and trend detection. Deploy a probability exponentially weighted moving average model. Continuously estimate the subsequent data scale according to the sensor data flow, update the model online and adjust the forwarding amount; it includes the following steps:

Step 2.1.1. Calculate the subsequent data volume changes based on the sensor data using the probability exponential weighting method: use a variable weighting factor to adapt to the impact of sudden and transient changes in the data:

为概率上可变化的加权因子，

P_i为第i次迭代的权重；

值是当前δ_i的概率；β是P_i的权重；α为加权因子，0<α<1；概率指数加权移动平均值模型的原理是当前δ_i对估计过程有概率为p的贡献。因此，将权重更新为1-βP_i，以便在估计过程中考虑突然却对后续估计几乎没有影响的意外峰值，从而限制模型过高估计后续δ_i；如果意外峰值之后是传感数据流的持续变化，则随后的意外峰值会被赋予更大的p值，从而允许它们对估计过程产生更大影响；In the above formula, _{u 1} is the current sensor data flow change, and δ _i is the distance between two consecutive data points vi and vi _-1 ;

is a probabilistically variable weighting factor,

P _i is the weight of the ith iteration;

The value is the probability of the current δ _i ; β is the weight of P _i ; α is the weighting factor, 0<α<1; the principle of the probability exponentially weighted moving average model is that the current δ _i contributes to the estimation process with probability p. Therefore, the weights are updated to 1-βP _i to account for unexpected peaks that are abrupt but have little effect on subsequent estimates during the estimation process, thereby limiting the model to overestimate subsequent δ _i ; if the unexpected peak is followed by a continuous stream of sensory data changes, subsequent unexpected peaks are assigned larger p-values, allowing them to have a greater impact on the estimation process;

Step 2.1.2. Although the probabilistic exponentially weighted moving average model avoids the model's overestimation of unexpected peaks, it does not take into account the monotonic phases of the upward and downward trends, which tend to introduce time lag effects in the estimation process. Therefore, the model proposed by Holt et al. is used to estimate monotonic growth or monotonic decay in sensory data flow changes:

In the above formula, x _i is the current sensing data flow; u ₁ is the current sensing data flow change; δ _i is the distance between two consecutive data points vi and v _{i -1} ; ξ is [0,1 ] in the range of smoothing weights, a value of ξ close to 1 indicates a preference for recent trends; in order to initialize the calculation of X ₁ , the initial value of i is set to 2; during the estimation process, the moving average of the probability exponentially weighted moving average model is used. value is raised to an appropriate numerical base to reduce hysteresis

Step 2.1.3: Deploy a data scale estimation model on each local device according to the above calculation results, estimate the subsequent sensing data scale according to the sensing data flow, update the probability exponentially weighted moving average model online and adjust the data forwarding amount.

Preferably, the step 2.2 specifically includes the following steps:

Step _2.2.1 . Monitor the estimated sampling period Ti ₊₁ of the next data point according to the current sampling period Ti: increase the estimated sampling period Ti ₊₁ of the next data point if the load decreases, and decrease if the load increases The next data point estimates the sampling period T _i+1 ; the magnitude of the increase or decrease of the next data point estimated sampling period T _i+1 depends on the confidence level c _i , indicating that the adaptive sampling model estimates and follows the sensory data flow The current changing confidence c _i , when the sensing data flow confidence c _i is larger, the adaptive sampling model adopts a larger sampling period;

Step 2.2.2. After updating the data scale estimation model and calculating the sensory data flow confidence ci, compare it with the user-defined imprecision γ, _γ∈ [0,1]; meanwhile, calculate the new sampling period T _i+1 :

In the above formula, λ is the weight coefficient, c _i is the confidence level; γ is the imprecision, if γ→0, the above formula will converge to the periodic sampling method; if γ→1, even if a reliable estimate cannot be made, the Adjustments are made every sampling interval; therefore, if the data size estimation model fails to provide an estimate within a certain confidence interval, the adaptive sampling model will roll back to the default sampling period _Tmin at the next data estimation point di ₊₁ .

Preferably, the step 2.3 specifically includes the following steps:

Step 2.3.1. Calculate the Fano factor over a period of time window:

In the above formula, σ ² is the variance, μ is the mean value, both σ _i and μ _i are weighted and calculated by the change probability P provided by the data scale estimation model; the Fano factor is expressed as W, as the ratio of the variance σ ² to the mean value μ ;

Step 2.3.2. After calculating F _i , compare the error variance σ _err with the maximum imprecision γ provided by the user: if F _i indicates that the current sensing data stream is not scattered and σ _err is less than γ, the filtering range becomes larger , filter out nearby values, while still keeping the data as a whole within the user-defined accuracy requirement; if _Fi indicates that the current flow of sensory data is overly dispersed, shorten the filter range or restore it to the default value and report anomalies in the data .

Preferably, the imprecise parameter γ in the step 2.2.2 is used to set the sensitivity.

Preferably, the time complexity of the adaptive sampling method of the adaptive sampling model in step 2.2 is constant time, because all calculations are based on pre-collected values and the entire sensory data flow information is not required.

The beneficial effects of the invention are as follows: the invention estimates the transmission flow based on the lightweight estimation model for the environmental parameter data collected by the sensor during the growth process of the famous flower, filters the sensing data flow through adaptive sampling and adaptive filtering technology, and uses fog The calculation enables real-time monitoring of the appropriate amount of effective sensor data.

Description of drawings

Figure 1 is a schematic diagram of the sensor network topology of the fine cultivation factory of famous flowers and precious medicinal plants;

Fig. 2 is the flow chart of the fog computing monitoring technology of the plant factory;

Figure 3 is a flowchart of a lightweight approximate sensor flow numerical estimation model;

FIG. 4 is a flowchart of adaptive sampling and adaptive filtering.

Detailed ways

The present invention will be further described below in conjunction with the embodiments. The following examples are illustrative only to aid in the understanding of the present invention. It should be pointed out that for those skilled in the art, without departing from the principle of the present invention, the present invention can also be modified several times, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

The invention processes the sensor data of flowers and medicinal plants based on the fog computing mode, and extends and expands the computing, network and storage capabilities of the cloud of the current plant factory monitoring system to the edge of the network.

1. The overall idea of the present invention:

The present invention mainly solves the following problems: how to solve the network bandwidth and delay bottleneck faced by the monitoring of environmental parameters of fine cultivation of famous flowers and medicinal plants in plant factories. On the one hand, this technology performs data processing locally, so that the computing time for data preprocessing is shorter and the bandwidth required is less, thereby reducing the network requirements for the monitoring system. The local end of the monitoring system using the fog computing mode needs to selectively filter, sample, filter and forward the sensor data. The sensor data is transmitted to the cloud server through the above processing, which requires less storage space and can more conveniently monitor and visualize the data.

2. Example

Example 1

A monitoring system for a famous flower and medicinal plant cultivation factory in an intelligent plant factory environment, comprising: a plant factory data acquisition module, a fog acquisition module, a cloud storage module and a monitoring system visualization module; the plant factory data acquisition module is connected through WiFi or Bluetooth The fog collecting module is connected for data transmission, the fog collecting module uploads data to the cloud storage module, and the monitoring system visualization module reads data from the cloud storage module. The plant factory data collection module is provided with sensors, and the sensors include a temperature sensor, a carbon dioxide sensor and a humidity sensor; the fog collection module is used for data scale estimation, adaptive sampling and adaptive filtering; the cloud storage module It is used for uploading data to the cloud and backing up the data locally; the monitoring system visualization module is used for environmental data visualization, plant growth visualization and nutrient solution status visualization.

Example 2

The fog computing monitoring technology of the monitoring system for famous flowers and medicinal plants cultivation factory under the intelligent plant factory environment of the present invention comprises the following steps, as shown in Figure 1:

A. Divide the plant factory into regions, and deploy several local devices in each region to collect sensing data such as temperature, humidity, carbon dioxide, etc.:

Step A includes the following steps:

Build a sensor network and collect relevant information, such as temperature and humidity, carbon dioxide, and soil pH, according to the monitoring strategy of the fog computing module. The monitoring area is divided for each local fog computing device, and the relevant data collected by all sensors in the monitoring area is transmitted to the local device for subsequent forwarding.

B. After each local device collects the sensing data, it estimates the subsequent data scale locally through a lightweight data scale estimation model to adjust the forwarding amount, and at the same time, the sensing data is automatically processed through the local adaptive sampling and filtering model. Adaptive filtering, and finally forward the processed sensor data to the cloud server with the adjusted forwarding amount through the forwarding device, as shown in Figure 2:

Step B includes the following steps:

B11. Deploy a lightweight sensing data scale estimation model on a local device, and use the monitoring data of a specific environmental monitoring device in the past period of time to predict its monitoring data in the future.

The distance δ _i between two consecutive data points v _i and v _i-1 values is defined as follows:

δ _i =|v _i -v _i-1 |

The distance δ _i is used to update the change of the local reference running time of the sensor data stream ρ(M), and the current sensor data stream change is calculated by moving average, which is expressed as μ _i . Denote the distance of the next two data points as δ _i+1 . Intuitively, the distance between two consecutive values represents a change in the sensory data stream. An exponentially weighted moving average model is used for approximate prediction of monitoring sensor data, and a weighting factor (0 < α < 1) is introduced to reduce the weight of older sensor data in an exponential change pattern, as shown in the following formula:

In the above formula, u ₁ is initialized with δ ₁ and updated iteratively later. Although the exponentially weighted moving average model is more suitable for the monitoring needs of actual plant factories, its response to transient changes is unstable, so it cannot always be assumed that only exponential weighting exists. Specifically, if an exponentially weighted moving average model encounters a sudden peak after a longer period of stabilization, and if this abrupt peak is followed by a stabilization period, the model retains the peak. This will lead to overestimation of the subsequent δ _i , which will affect the accuracy of the approximate estimation model.

In order to solve the problem that the above model encounters a sudden data peak change, the model is solved by two steps of probability weighting and trend detection, and the model is deployed on each local device. The subsequent data scale is continuously estimated based on the sensor data flow, the model is updated online and the forwarding amount is adjusted. Step B11 specifically includes the following three steps:

B111. Use a variable weighting factor to adapt to the effects of sudden transient changes in the data, as follows:

在上式中，P_i为第i次迭代的权重，

值是当前δ_i的概率，遵循传感数据流变化的建模分布，β是P_i的权重。概率指数加权移动平均值模型的原理是当前δ_i对估计过程有概率为p的贡献。因此，将权重更新为1-βP_i，以便在估计过程中考虑突然却对后续估计几乎没有影响的“意外”峰值，从而限制模型过高估计后续δ_i。如果“意外”峰值之后是传感数据流的持续变化，则随后的“意外”峰值会被赋予更大的p值，从而允许它们对估计过程产生更大影响。The probabilistic exponentially weighted moving average model introduces probabilistically variable weighting factors

In the above formula, Pi is the weight of the _ith iteration,

The value is the probability of the current δ _i , following the modeled distribution of sensory data flow changes, and β is the weight of P _i . The principle of the probability exponentially weighted moving average model is that the current δ _i has a probability p contribution to the estimation process. Therefore, the weights are updated to 1-βP _i to account for sudden "unexpected" peaks in the estimation process that have little effect on subsequent estimates, thereby limiting the model to overestimate subsequent δ _i . If the "unexpected" peak is followed by a persistent change in the sensory data stream, subsequent "unexpected" peaks are assigned larger p-values, allowing them to have a greater impact on the estimation process.

B112. Although the probability exponentially weighted moving average model avoids the model's overestimation of unexpected peaks, it does not take into account the monotonic phases of up and down trends, which tend to introduce time lag effects in the estimation process. Therefore, the monotonic growth/decay in metric flow variation is estimated using the model proposed by Holt et al. as follows:

where ξ is a smoothing weight in the range [0,1], where values close to 1 indicate a preference for recent trends. The initial value of i is set to 2 to initialize the calculation of X ₁ . Therefore, in the estimation process, the hysteresis effect is reduced by raising the moving average of the probability exponentially weighted moving average model to an appropriate numerical base.

B113. Combine the above two steps, and deploy the model on each local device. The subsequent data scale is continuously estimated based on the sensor data flow, the model is updated online and the forwarding amount is adjusted.

B12. Deploy an adaptive sampling model for fine cultivation monitoring in a plant factory on a local device, sample sensor data and adjust the sampling period, as shown in Figure 3. The core of adaptive sampling is the process of dynamically adjusting the sampling periodicity T _i based on changes in the monitoring sensor data flow of the plant factory, while the monitoring accuracy still meets the accuracy requirements given by the user.

The estimated sampling period Ti ₊₁ for monitoring the next data point depends on the current sampling period _Ti . If the load decreases, the estimated sampling period for the next data point is increased, and if the load increases, the estimated sampling period for the next data point is decreased in turn. cycle. The magnitude of the need to increase or decrease depends on the confidence _ci , which represents the confidence that the model estimates and follows the current change in the metric flow. When the sensory data flow estimation model is more "confident" (confidence c _i is larger), adaptive sampling will use a larger sampling period. Therefore, compared to the threshold sampling technique that adjusts the sampling rate only based on the data point values _vi , the proposed adaptive sampling considers both the sensory data flow variation and the confidence of the model estimation.

After updating the estimated model and computing the estimated confidence, it is compared to an acceptable user-defined imprecision (denoted γ). The imprecise parameter (γ∈[0,1]) is used to set the sensitivity while calculating the new sampling period T _i+1 as follows:

In the above formula, λ is the weight coefficient. If γ→0, the above formula will converge to the periodic sampling method. Conversely, if γ→1, every sampling interval produces an adjustment even if a reliable estimate cannot be made. Therefore, if the sensor flow data estimation model cannot provide an estimate within a certain confidence interval, the adaptive sampling algorithm will roll back to the default sampling period _Tmin at the next data estimation point d _i+1 . Furthermore, compared to the stepwise technique that adjusts the sampling rate only based on a step function (eg, T _i+1 ←T _i ±T _step ), T _step is the confidence interval radius; the proposed adaptive algorithm can The metric flow of , within the appropriate range [T _min , T _max ], responds efficiently.

The time complexity of the proposed adaptive sampling method is constant time because all computations are based on pre-collected values and the entire sensory data flow information is not required. The imprecision γ is the only parameter that is self-defined in the estimation process.

B13. Deploy an adaptive filtering model for fine cultivation monitoring of plant factories on the local device, filter the sampled sensor data, and adjust the sampling period according to the filtering results, as shown in Figure 4. Plant factory monitoring sensing data stream filtering is the process of suppressing data points when the continuous data point value is less than a range, denoting the suppression range as R. Therefore, if v _i ∈ [v _i-1 -R,v _i-1 +R], then filter the current data point d _{i of value v i .} However, this requires that the user previously know the data point value distribution and that this distribution will not change, otherwise there is no guarantee that any values will be filtered.

Through adaptive filtering technology, the filtering range R is dynamically adjusted according to changes in the monitored sensor data stream, while still meeting user-defined accuracy requirements. The filtering range R is determined by monitoring changes in the sensory stream data. At this time, only a small part of the data needs to be limited in accuracy, and R is gradually adjusted based on the number of previously filtered data points.

The Fano factor is used to show the degree of change associated with the current change in the monitored sensory data stream. The Fano factor (F ≥ 0), like the dispersion index, is a normalized measure of the dispersion of a probability distribution that quantifies whether a set of data points is clustered (F < 1) or dispersed compared to a statistical model. The Fano factor is computed over a window of time, ^denoted W, as the ratio of the variance σ to the mean μ, as follows:

The variance σ ² and mean μ of the above formula do not require additional calculations, because σ _i and μ _i are both weighted by the change probability P provided by the sensor data flow estimation model; at the same time, there is no need to use the data point window, because σ _i and _μi are weighted according to previous values to adjust the contribution of each data point accordingly. Intuitively, when σ _i decreases _, the Fano factor Fi follows the change, indicating a decrease in the size of the data stream. After _{Fi is calculated, σ err (} error variance) is compared to the user-supplied maximum tolerable imprecision, denoted γ. If _Fi indicates that the sensory data stream is not scattered and σ _err is less than γ, the filtering range becomes larger, trying to filter out nearby values, while still keeping the data as a whole within the user-defined accuracy requirement. Otherwise, if Fi _indicates that the current flow of sensory data is excessively scattered, shorten the filtering range or restore it to the default value and report anomalies in the data.

Like adaptive sampling, the adaptive filtering algorithm has a time and space complexity of O(1) because R _i+1 is computed from its previous value, while μ _i , σ _i and σ _err are run-time estimates of the model by output as previously described.

B14. Deploy an adaptive forwarding model for fine cultivation monitoring of plant factories on the local device, forward the sampled and filtered sensor data according to the forwarding amount estimated by the sensor data scale estimation model, and transmit the data to the cloud server . Through the fog computing model, the forwarding process requires less bandwidth and thus reduces the demand on the network.

C. The cloud server receives and stores the sensor data, and at the same time displays important data such as plant growth status data, plant growth environment data and other important data on the device, providing real-time data monitoring to managers:

Step C includes the following steps:

Receive and parse the plant factory sensor data forwarded by the local device through the cloud server, and save different types of sensor data to the local database.

Through suitable visualization charts, various sensor data information of the plant factory can be displayed in real time.

3. Experimental conclusion:

It can be seen that the system can estimate the subsequent data scale according to the data scale collected by the sensor to adjust the forwarding amount, and perform adaptive sampling and filtering according to the current data, so as to forward the sensor data with an appropriate scale and low noise to the cloud server. . While storing data, the cloud server can also display real-time plant growth status data and plant growth environment data. In this way, cultivators can also achieve a more accurate understanding of the greenhouse environment under the condition of remote monitoring, and the method is universal and can be applied to a variety of crops.

Claims (7)

1. A fog computing monitoring technology for a monitoring system for famous flowers and medicinal plants cultivation factories in an intelligent plant factory environment, the monitoring system comprising: a plant factory data acquisition module, a fog acquisition module, a cloud storage module and a monitoring system visualization module; the The plant factory data collection module is connected to the fog collection module through WiFi or Bluetooth for data transmission, the fog collection module uploads data to the cloud storage module, and the monitoring system visualization module reads data from the cloud storage module; it is characterized in that the fog computing monitoring The technique includes the following steps: Step 1. Build a sensor network, divide the plant factory into areas, and divide the monitoring area for each local fog computing device. The plant factory data acquisition module deploys several local devices in each area to collect sensory data; The sensor data is uploaded to the fog collection module; Step 2. After each local device collects the sensor data, it estimates the subsequent data scale locally through the lightweight data scale estimation model to adjust the forwarding amount, and at the same time uses the local adaptive sampling model and adaptive filtering model to adjust the forwarding amount. The sensor data is processed, and finally the processed sensor data is forwarded to the cloud server through the forwarding device according to the adjusted forwarding amount; Step 2.1. According to the scale of the sensing data, estimate the subsequent data scale through the probability exponentially weighted moving average model and trend detection, update the probability exponentially weighted moving average model online and adjust the amount of data forwarding: _{Deploy a} lightweight data scale estimation model on the local device, and use the monitoring data of a specific environmental monitoring device in the past period to predict its future monitoring data; The distance δ _i is defined as follows: δ _i =|v _i -v _i-1 | Update the change of the local reference running time of the sensing data stream ρ(M) with the distance δ _i , calculate the current change of the sensing data stream by moving average, denoted as μ _i , and denote the distance of the next two data points as δ _i+1 ; use the probability exponential weighted moving average model to approximate the monitoring sensor data, and introduce a weighting factor (0<α<1) to reduce the weight of the sensor data according to the exponential change mode, as shown in the following formula:

In the above formula, use δ ₁ to initialize u ₁ ; Step 2.2. Deploy the adaptive sampling model for fine cultivation monitoring of plant factories on the local device, and calculate the confidence level according to the scale of the sensor data. After comparing with the imprecision defined by the user, adjust the sampling period and sample the sensor data. ; Step 2.3. Deploy the adaptive filtering model for fine cultivation monitoring of plant factories on the local device, and filter the sampled sensor data: denote the filtering range as R, if the value v _i ∈ [v _i-1 -R ,v _i-1 +R], then filter the current data point d _i of v _i ; And according to Fano factor, according to monitoring sensor data flow change, dynamically adjust filtering range R, adjust sampling period according to filtering result; Step 2.4, deploying an adaptive forwarding model for fine cultivation monitoring of plant factories on the local device, forwarding the sampled or filtered data according to the forwarding amount estimated by the data scale estimation model, and transmitting the data to the cloud server; Step 3. The cloud server receives and parses the sensory data forwarded by the local device, saves different types of sensory data to the local database, and displays the plant growth status data or plant growth environment data on the device through a visual chart, to the management personnel. Provides real-time data monitoring. 2. The fog computing monitoring technology of the monitoring system for famous flowers and medicinal plants cultivation factory under the intelligent plant factory environment according to claim 1, is characterized in that: in the described step 1, the local device comprises a sensor. 3. according to the fog computing monitoring technology of famous flower and medicinal plant cultivation factory monitoring system under the intelligent plant factory environment of claim 1, it is characterized in that, described step 2.1 specifically comprises the following steps: Step 2.1.1. Calculate subsequent data volume changes based on the sensor data using the probability exponential weighting method: Use weighting factors to adapt to the impact of sudden transient changes in data:

In the above formula, _{u 1} is the current sensor data flow change, and δ _i is the distance between two consecutive data points vi and vi _-1 ;

is the weighting factor,

P _i is the weight of the ith iteration;

The value is the probability of the current δ _i ; β is the weight of P _i ; α is the weighting factor, 0<α<1; update the weight to 1-βP _i ; Step 2.1.2. Use the model proposed by Holt to estimate the monotonic growth or monotonic decay in the sensory data flow change:

In the above formula, x _i is the current sensing data flow; u ₁ is the current sensing data flow change; δ _i is the distance between two consecutive data points vi and v _{i -1} ; ξ is [0,1 ] The smoothing weight in the range; the initial value of i is set to 2; Step 2.1.3: Deploy a data scale estimation model on each local device according to the above calculation results, estimate the subsequent sensing data scale according to the sensing data flow, update the probability exponentially weighted moving average model online and adjust the data forwarding amount. 4. according to the fog computing monitoring technology of famous flower and medicinal plant cultivation factory monitoring system under the intelligent plant factory environment of claim 1, it is characterized in that, described step 2.2 specifically comprises the steps: Step _2.2.1 . Monitor the estimated sampling period Ti ₊₁ of the next data point according to the current sampling period Ti: increase the estimated sampling period Ti ₊₁ of the next data point if the load decreases, and decrease if the load increases The next data point estimated sampling period T _i+1 ; the magnitude of the increase or decrease of the next data point estimated sampling period T _i+1 depends on the confidence c _i , when the sensing data flow confidence c _i is large , the adaptive sampling model adopts a larger sampling period; Step 2.2.2. After updating the data scale estimation model and calculating the sensory data flow confidence ci, compare it with the user-defined imprecision γ, _γ∈ [0,1]; meanwhile, calculate the new sampling period T _i+1 :

In the above formula, λ is the weight coefficient, _ci is the confidence level; γ is the imprecision, if γ→0, the above formula will converge to the periodic sampling method; if γ→1, adjust each sampling interval; if If the data size estimation model cannot provide an estimate within a certain confidence interval, the adaptive sampling model will roll back to the default sampling period T _min at the next data estimation point d _i+1 . 5. according to the fog computing monitoring technology of famous flower and medicinal plant cultivation factory monitoring system under the intelligent plant factory environment of claim 1, it is characterized in that, described step 2.3 specifically comprises the steps: Step 2.3.1. Calculate the Fano factor over a period of time window:

In the above formula, σ ² is the variance, μ is the mean value, both σ _i and μ _i are weighted and calculated by the change probability P provided by the data scale estimation model; the Fano factor is expressed as W, as the ratio of the variance σ ² to the mean value μ ; Step 2.3.2. After calculating F _i , compare the error variance σ _err with the maximum imprecision γ provided by the user: if F _i indicates that the current sensing data stream is not scattered and σ _err is less than γ, the filtering range becomes larger , filter out nearby values; if _Fi indicates that the current flow of sensory data is excessively scattered, shorten the filtering range or restore it to the default value and report anomalies in the data. 6. The fog computing monitoring technology of the monitoring system for famous flowers and medicinal plants cultivation factory under the environment of the intelligent plant factory according to claim 4, characterized in that: in the step 2.2.2, the imprecise parameter γ is used to set the sensitivity. 7. according to the fog computing monitoring technology of famous flowers and medicinal plant cultivation factory monitoring system under the intelligent plant factory environment of claim 1, it is characterized in that: the time complexity of the self-adaptive sampling method of self-adaptive sampling model in described step 2.2 is constant time.

Images