Intelligent control-based hermetia illucens fecal sewage treatment system

Abstract

The application discloses a black soldier fly fecal treatment system based on intelligent control and relates to the technical field of environmental protection, which comprises a central control unit that controls the release of a preset number of larvae and sends the number to a treatment capacity prediction module; a sensor module is connected with the central control unit to collect temperature and humidity data, a preset temperature and humidity range is set in the central control unit, the temperature and humidity range is used to adjust the temperature and humidity for the growth of the larvae, and the adjusted temperature and humidity are sent to the treatment capacity prediction module; a disease detection module monitors the color change of the larvae in real time, calculates a disease index of the larvae, and sends the disease index and image data to the treatment capacity prediction module through the central control unit; the treatment capacity prediction module predicts the fecal treatment capacity based on the disease index, temperature and humidity and number of the larvae, and the central control unit releases feces in the next release cycle based on the initial fecal release amount and the fecal treatment capacity. The application evaluates the fecal treatment capacity of the larvae by the temperature and humidity, the number of the larvae and the disease condition of the larvae, and realizes automatic fecal treatment.

Description

Technical Field

[0001] This invention relates to the field of environmental protection technology, and more specifically to a black soldier fly larvae waste treatment system based on intelligent control. Background Technology

[0002] Black soldier flies (Hermetia illucens) are insects with broad application prospects in waste treatment and organic waste degradation. Black soldier fly larvae can efficiently convert organic waste into biomass protein and fertilizer, thus attracting increasing attention in agriculture, animal husbandry, and waste management. Currently, many farms and organic waste treatment facilities are beginning to introduce black soldier flies to treat organic waste such as manure. Traditional black soldier fly manure treatment technology mainly relies on manual management, which limits treatment efficiency and capacity, and also has high requirements for environmental conditions (such as temperature and humidity).

[0003] Although black soldier fly larvae have shown some success in treating sewage, existing technologies have several significant shortcomings. First, the level of automation in the treatment process is low, typically relying on manual monitoring and operation, making large-scale, high-efficiency treatment difficult. Second, the monitoring and control of the black soldier fly larvae's growth environment is rather rudimentary, leading to significant fluctuations in treatment efficiency, especially under unstable environmental conditions such as temperature and humidity, where the treatment capacity may decrease significantly. Furthermore, existing systems lack intelligent management tools for waste disposal, process monitoring, and sewage separation, easily resulting in resource waste and inefficiency.

[0004] Therefore, how to automate the treatment of black soldier fly larvae is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention provides a black soldier fly larvae frass treatment system based on intelligent control, which accurately assesses the amount of larval frass to be treated by taking into account temperature and humidity, larval number, and larval disease status, thereby achieving high efficiency in frass treatment and enabling automatic frass treatment.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A black soldier fly larvae treatment system based on intelligent control includes: a central control unit, a temperature and humidity control module, a larvae dispensing module, a sensor module, a larvae dispensing module, a disease detection module, and a treatment volume prediction module. The central control unit controls the larvae dispensing module to dispense a preset number of larvae and sends the data to the treatment volume prediction module. The sensor module includes a temperature sensor and a humidity sensor, which are connected to the central control unit. The central control unit has a preset temperature and humidity range. Based on this preset range, the temperature and humidity control module adjusts the temperature and humidity for larval growth and sends the data to the treatment volume prediction module. The disease detection module monitors changes in larval body color in real time, calculates a larval disease index, and sends this data, along with image data, to the treatment volume prediction module via the central control unit. The treatment volume prediction module predicts the larvae treatment volume based on the disease index, larval growth temperature and humidity, and the number of larvae, and sends this prediction to the central control unit. The central control unit dispenses larvae in the next dispensing cycle based on the initial larvae dispensing volume and the predicted treatment volume.

[0008] Preferably, the disease detection module specifically includes:

[0009] The feature value extraction module extracts the feature values ​​of the disease at each collection moment within a set time period;

[0010] The disease index calculation module determines the disease index based on the disease's characteristic values; the formula for determining the disease index is as follows:

[0011]

[0012] Among them, F t w represents the disease index at time t. j f represents the weight corresponding to the j-th disease characteristic value. j Let j represent the characteristic value of the j-th disease, where j = 1, 2, ..., m.

[0013] Preferably, the processing volume prediction module specifically includes:

[0014] The model building module utilizes a deep convolutional neural network and a temporal information storage network to integrate the larval growth temperature and humidity, disease index, and environmental attributes of black soldier fly frass at different times during frass treatment, and constructs a frass treatment volume prediction model.

[0015] The model training module, which involves the joint training of a time-series information storage network and a deep convolutional neural network, uses historical corresponding larval release quantity image data as training samples for the deep convolutional neural network and larval growth temperature and humidity and disease index as training samples for the time-series information storage network, and performs joint training of the two.

[0016] The model prediction module obtains the temperature and humidity of the larvae's growth, the disease index, and the number of larvae. It then obtains the corresponding manure treatment volume prediction distribution through the manure treatment volume prediction model and selects the treatment volume with the highest probability of exceeding a preset threshold as the manure treatment volume corresponding to the test data.

[0017] Preferably, the model building module specifically includes:

[0018] The joint learning module models the features of black soldier fly frass treatment data, and selects the larval growth temperature and humidity, disease index and larval number that affect the black soldier fly frass treatment amount for temporal joint learning;

[0019] The prediction distribution module stores the temporal information from multiple iterations into the final hidden layer state h of the network unit. t The data is passed as input to the output layer, and the probability distribution y of the sewage treatment volume is estimated using the softmax function. t The formula is as follows:

[0020] y t =softmax(W*h t +b);

[0021] Where W and b are the weights and biases, respectively.

[0022] Preferably, the model training module specifically includes:

[0023] The initialization module performs end-to-end training of the deep convolutional neural network, pre-training the model using image data and initializing it.

[0024] The vector value calculation module trains the temporal information storage network by performing forward computation. It uses a pre-trained deep convolutional neural network to extract features from image data obtained at different time periods, and sequentially calculates the output value of each neuron through the temporal information storage network. The temporal information storage network then calculates i... t c t f t o t h t The values ​​of the five vectors;

[0025] The weight gradient calculation module simultaneously fine-tunes all parameters of the deep convolutional neural network and the temporal information storage network, and calculates the error term δ value of each neuron in reverse. The backpropagation of the error term of the temporal information storage network includes two directions: one is backpropagation along time, starting from the current time t, to calculate the error term at each time step; the other is to propagate the error term to the upper layer of the feature extraction network, and calculate the gradient of each weight based on the corresponding error term.

[0026] Preferably, the joint learning module specifically includes:

[0027] The attribute setting module allows you to define time-invariant environmental attribute elements, where K and L represent the black soldier fly species and feces type, respectively; and time-varying attribute elements, such as the temperature C, which varies with time from t=1 to t=T. t Humidity S t and disease index F t ;

[0028] The rules for updating the time-series information storage network are set in the rule setting module as follows:

[0029] x0 = T1 * A1;

[0030] x t =T2*{F,A2} t ,t∈{1,...,T};

[0031] Where A1 is the time-invariant environmental attribute element {K,L}, and A2 is the time-variant attribute element {C}. t ,S t ,F t Let T1 and T2 be the transformation matrices for time-invariant and time-varying attributes, respectively. The feature set of time-invariant environmental attribute factors of the time-series information storage network is {x0, x1, ..., x...}. T};

[0032] The forward computation setting module normalizes the feature data, controlling the data dimension between 0 and 1. It sets the number of layers in the time-series information storage network model to be consistent with the time points, and each layer has inputs and outputs. The first layer of the time-series information storage network model takes the time-invariant environmental attribute factor feature x0 as input, and its influence is continuously passed to the prediction at each moment. The input of the (n+1)th layer of the time-series information storage network model is the image after being processed by the deep feature extraction network and fused with the time-varying environmental attribute factor, which together serve as the input of the (n+1)th layer of the time-series information storage network model. The layers of the time-series information storage network model are input sequentially according to the order of different time-series shooting, setting the forward computation of the time-series information storage network unit model.

[0033] Preferably, the forward calculation setting module specifically includes:

[0034] Long-term storage unit c t-1 Through the forgetting unit f t To discard information and set up a forgetting unit f t Subject to the external input x at the current time t The short-time storage output h from the previous time step t-1 Long-term storage of the previous time step c t-1 The control is expressed as follows:

[0035] f t =σ(W xf x t +W hf h t-1 +W cf c t-1 +b f );

[0036] W xf W hf W cf b f These represent the weights of the external input, the weights stored in the short time, the weights stored in the long time, and the bias, respectively.

[0037] The external input x at the current time t and the short-time storage output h from the previous time step t-1 Calculate the new information at the current moment

[0038]

[0039] From input unit i t Control will use the new information at the current moment Write to a long-term storage unit to generate a new long-term storage c. t Its expression is as follows:

[0040] i t =σ(W xi x t +W hi h t-1 +W ci c t-1 +b i );

[0041]

[0042] Among them, i t Subject to x t h t-1 c t-1 control, i t c t f t o t These are the input unit, long-term storage unit, forgetting unit, and output unit of the time-series information storage network unit, respectively; where σ(·) represents the sigmoid activation function, · represents component multiplication, and W is the weight matrix connecting different units;

[0043] Activate long-term storage unit c t , by output unit o t Control the accumulated storage c tSelecting some relevant storage components generates the storage h that is of interest at this moment. t Then output this part of the storage as y. t o t =σ(W xo x t +W ho h t-1 +W co c t-1 +b o ), h t =o t ·tanh(c t );

[0044] Among them, the output unit o t Subject to x t h t-1 and the long-term storage c at the current moment t Control.

[0045] As can be seen from the above technical solution, compared with the prior art, this invention discloses a black soldier fly manure treatment system based on intelligent control. The system achieves intelligent control of the entire system through a central control unit, including larval release, temperature and humidity control, and manure disposal, thus improving the system's automation level and operational efficiency. Utilizing black soldier flies to treat manure, combined with modules for temperature and humidity control, disease detection, and treatment volume prediction, can more effectively treat manure and reduce environmental pollution. The treatment volume prediction module predicts based on multiple factors such as disease index, larval growth temperature and humidity, and larval number, improving the accuracy of manure treatment volume prediction and helping to optimize manure disposal strategies. The sensor module and disease detection module can monitor the larval growth environment and health status in real time. The central control unit adjusts system parameters promptly based on this information to ensure the growth of black soldier flies and the effectiveness of manure treatment. The joint training of a temporal information storage network and a deep convolutional neural network can fully integrate multiple temporal attribute elements of black soldier fly manure treatment, improving model performance and predictive ability. Adjustments and optimizations can be made according to different environmental attributes and black soldier fly species to adapt to different application scenarios and needs. Attached Figure Description

[0046] 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, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0047] Figure 1 This is a schematic diagram of the structure provided by the present invention;

[0048] Figure 2 This is a schematic diagram of the joint learning module provided by the present invention. Detailed Implementation

[0049] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] This invention discloses a black soldier fly dung treatment system based on intelligent control, such as... Figure 1 As shown, the system includes: a central control unit, a temperature and humidity control module, a manure dispensing module, a sensor module, a larvae dispensing module, a disease detection module, and a processing capacity prediction module. The central control unit controls the larvae dispensing module to dispense a preset number of larvae and sends the data to the processing capacity prediction module. The sensor module includes a temperature sensor and a humidity sensor, which are connected to the central control unit. The central control unit has a preset temperature and humidity range. Based on the preset temperature and humidity range, the temperature and humidity control module adjusts the temperature and humidity for larval growth and sends the data to the processing capacity prediction module. The disease detection module monitors changes in larval body color in real time, calculates the larval disease index, and sends the data, along with image data, to the processing capacity prediction module through the central control unit. The processing capacity prediction module predicts the manure processing capacity based on the disease index, larval growth temperature and humidity, and the number of larvae, and sends the prediction to the central control unit. The central control unit dispenses manure in the next dispensing cycle based on the initial manure dispensing volume and the manure processing capacity.

[0051] In one specific embodiment, the disease detection module employs either spectral analysis technology or machine learning algorithms, both of which are existing technologies. A spectral sensor is used to illuminate the larvae, acquiring spectral information of their reflection or emission. Different body colors lead to changes in spectral characteristics; by analyzing the spectral data, changes in larval body color can be monitored in real time. Combining computer vision technology, machine learning algorithms, such as convolutional neural networks (CNNs) in deep learning, are used to train a large number of larval images, enabling the model to automatically identify and judge changes in larval body color.

[0052] In one specific embodiment, the disease detection module specifically includes:

[0053] The feature value extraction module extracts the feature values ​​of the disease at each collection moment within a set time period;

[0054] The disease index calculation module determines the disease index based on the disease's characteristic values; the formula for determining the disease index is:

[0055]

[0056] Among them, F t w represents the disease index at time t. j f represents the weight corresponding to the j-th disease characteristic value. j Let j represent the characteristic value of the j-th disease, where j = 1, 2, ..., m.

[0057] In one specific embodiment, the processing volume prediction module specifically includes:

[0058] The model building module utilizes a deep convolutional neural network and a temporal information storage network to integrate the larval growth temperature and humidity, disease index, and environmental attributes of black soldier fly frass at different times during frass treatment, and constructs a frass treatment volume prediction model.

[0059] The model training module, which involves the joint training of a time-series information storage network and a deep convolutional neural network, uses historical corresponding larval release quantity image data as training samples for the deep convolutional neural network and larval growth temperature and humidity and disease index as training samples for the time-series information storage network, and performs joint training of the two.

[0060] The model prediction module obtains the temperature and humidity of the larvae's growth, the disease index, and the number of larvae. It then obtains the corresponding manure treatment volume prediction distribution through the manure treatment volume prediction model and selects the treatment volume with the highest probability of exceeding the preset threshold as the manure treatment volume corresponding to the test data.

[0061] In one specific embodiment, the model building module specifically includes:

[0062] The joint learning module models the features of black soldier fly frass treatment data, and selects the larval growth temperature and humidity, disease index and larval number that affect the black soldier fly frass treatment amount for temporal joint learning;

[0063] The prediction distribution module stores the temporal information from multiple iterations into the final hidden layer state h of the network unit. t The data is passed as input to the output layer, and the probability distribution y of the sewage treatment volume is estimated using the softmax function. t The formula is as follows:

[0064] y t =softmax(W*h t +b);

[0065] Where W and b are the weights and biases, respectively.

[0066] By integrating the temperature and humidity of black soldier fly larvae growth, disease index, and environmental attributes at different times during black soldier fly frass treatment, a predictive model for frass treatment volume was constructed. Convolutional neural networks are highly adaptable and well-suited for processing data with statistical stationarity and local correlations, implicitly learning features of different shaped regions from captured training image data. Temporal information storage networks have significant advantages in learning long-term dependencies and temporal sequences in higher-level feature sequences. Therefore, combining the two models allows for temporal and spatial modeling of frass treatment volume.

[0067] In one specific embodiment, the model training module specifically includes:

[0068] The initialization module performs end-to-end training of the deep convolutional neural network, pre-training the model using image data and initializing it.

[0069] The vector value calculation module trains the temporal information storage network by performing forward computation. It uses a pre-trained deep convolutional neural network to extract features from image data obtained at different time periods, and sequentially calculates the output value of each neuron through the temporal information storage network. The temporal information storage network then calculates i... t c t f t o t h t The values ​​of the five vectors;

[0070] The weight gradient calculation module simultaneously fine-tunes all parameters of the deep convolutional neural network and the temporal information storage network, and calculates the error term δ value of each neuron in reverse. The backpropagation of the error term of the temporal information storage network includes two directions: one is backpropagation along time, starting from the current time t, to calculate the error term at each time step; the other is to propagate the error term to the upper layer of the feature extraction network, and calculate the gradient of each weight based on the corresponding error term.

[0071] In one specific embodiment, such as Figure 2 As shown, the joint learning module specifically includes:

[0072] The attribute setting module allows you to define time-invariant environmental attribute elements, where K and L represent the black soldier fly species and feces type, respectively; and time-varying attribute elements, such as the temperature C, which varies with time from t=1 to t=T. t Humidity S t and disease index F t ;

[0073] The rules for updating the time-series information storage network are set in the rule setting module as follows:

[0074] x0 = T1 * A1;

[0075] x t =T2*{F,A2} t ,t∈{1,...,T};

[0076] Where A1 is the time-invariant environmental attribute element {K,L}, and A2 is the time-variant attribute element {C}. t ,S t ,F t Let T1 and T2 be the transformation matrices for time-invariant and time-varying attributes, respectively. The feature set of time-invariant environmental attribute factors of the time-series information storage network is {x0, x1, ..., x...}. T}; F is the disease index set.

[0077] The forward computation module normalizes the feature data, controlling the data dimensionality between 0 and 1. Since the dimensions of several parameters in the feature data differ, data normalization is necessary to convert it into a dimensionless expression, which helps eliminate the influence of dimensions. The number of layers in the time-series information storage network model is set to be consistent with the time points, and each layer has inputs and outputs. Here, each layer of the time-series information storage network model corresponds to the stage process of a healthy black soldier fly developing into a diseased black soldier fly. The first layer of the time-series information storage network model inputs the time-invariant environmental attribute factor feature x0, and its influence is continuously passed to the prediction at each moment. This allows the prediction of the second layer to be influenced by the first layer, and the prediction of the third layer to be influenced by the second layer. Simultaneously, the time-series information storage network model includes a storage unit that attempts to store information for a longer period. According to the time sequence, the influence of all images and environmental attributes can be accumulated and stored sequentially, which is beneficial for the final prediction. The input to the (n+1)th layer of the temporal information storage network model is the image after passing through the deep feature extraction network and being fused with the temporal change environmental attribute factors. This is then used as the input to the (n+1)th layer of the temporal information storage network model, and the layers are sequentially input according to the order of different time-series shooting to set the forward computation of the temporal information storage network unit model.

[0078] In one specific embodiment, the forward calculation setting module specifically includes:

[0079] Long-term storage unit c t-1 Through the forgetting unit f t To discard information and set up a forgetting unit f t Subject to the external input x at the current time t The short-time storage output h from the previous time step t-1 Long-term storage of the previous time step c t-1 The control is expressed as follows:

[0080] ft =σ(W xf x t +W hf h t-1 +W cf c t-1 +b f );

[0081] W xf W hf W cf b f These represent the external input weights, short-term storage weights, long-term storage weights, and bias of the forgetting unit, respectively.

[0082] The external input x at the current time t and the short-time storage output h from the previous time step t-1 Calculate the new information at the current moment

[0083]

[0084] Among them, W xc Weights for external inputs of new information, W hc For short-sighted storage weight of new information, b c Bias for long-term storage of new information.

[0085] From input unit i t Control will use the new information at the current moment Write to a long-term storage unit to generate a new long-term storage c. t Its expression is as follows:

[0086] i t =σ(W xi x t +W hi h t-1 +W ci c t-1 +b i );

[0087]

[0088] Among them, i t Subject to x t h t-1 c t-1 control, i t c t f t o tThese represent the input unit, long-term storage unit, forgetting unit, and output unit of the time-series information storage network; where σ(·) represents the sigmoid activation function, · represents component multiplication, and W is the weight matrix connecting different units; W xi External input weights for the input unit, W hi For the short-sighted storage weights of the input unit, b i Long-term storage bias for the input unit.

[0089] Activate long-term storage unit c t , by output unit o t Control the accumulated storage c t Selecting some relevant storage components generates the storage h that is of interest at this moment. t Then output this part of the storage as y. t o t =σ(W xo x t +W ho h t-1 +W co c t-1 +b o ), h t =o t ·tanh(c t ); t is the current time, t-1 is the previous time.

[0090] Among them, the output unit o t Subject to x t h t-1 and the long-term storage c at the current moment t Control; W xo External input weights for the output unit, W ho For the short-sighted storage weight of the output unit, b o This is the long-term storage bias for the output unit.

[0091] In one specific embodiment, the preset temperature and humidity range within the central control unit specifically includes:

[0092]

[0093] Among them, Q opt For the optimal temperature setpoint, H opt For the optimal humidity setting, Q 当前 and H 当前 Given the current temperature and humidity, Q 目标 and H 目标 This is the optimal temperature and humidity range for black soldier flies.

[0094] In one specific embodiment, a communication module is also included. This module allows managers to monitor the processing in real time via a remote terminal and receive timely alarm information in case of anomalies. This feature makes system management more convenient and flexible, making it particularly suitable for large-scale farms or remote operation scenarios. This advantage stems from the combination of the communication module and the central control module, which enables wireless data transmission and remote monitoring.

[0095] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0096] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A black soldier fly dung treatment system based on intelligent control, characterized in that, include: The system comprises a central control unit, a temperature and humidity control module, a manure dispensing module, a sensor module, a larvae dispensing module, a disease detection module, and a treatment volume prediction module. The central control unit controls the larvae dispensing module to release a preset number of larvae and sends the data to the treatment volume prediction module. The sensor module includes a temperature sensor and a humidity sensor, both connected to the central control unit. The central control unit has a preset temperature and humidity range. Based on this range, the temperature and humidity control module adjusts the temperature and humidity for larval growth and sends the data to the treatment volume prediction module. The disease detection module monitors larval color changes in real time, calculates the larval disease index, and sends this data, along with image data, to the treatment volume prediction module via the central control unit. The treatment volume prediction module predicts the manure treatment volume based on the disease index, larval growth temperature and humidity, and the number of larvae, and sends this prediction to the central control unit. The central control unit then dispenses manure in the next dispensing cycle based on the initial manure dispensing volume and the predicted treatment volume. The disease detection module specifically includes: The feature value extraction module extracts the feature values ​​of the disease at each collection moment within a set time period; The disease index calculation module determines the disease index based on the disease's characteristic values; the formula for determining the disease index is as follows: ； in, express t Disease index at all times. Indicates the first j The weights corresponding to the characteristic values ​​of the diseases Indicates the first j Characteristic values ​​of the disease, ; The processing volume prediction module specifically includes: The model building module utilizes a deep convolutional neural network and a temporal information storage network to integrate the larval growth temperature and humidity, disease index, and environmental attributes of black soldier fly frass at different times during frass treatment, and constructs a frass treatment volume prediction model. The model training module, which involves the joint training of a time-series information storage network and a deep convolutional neural network, uses historical corresponding larval release quantity image data as training samples for the deep convolutional neural network and larval growth temperature and humidity and disease index as training samples for the time-series information storage network, and performs joint training of the two. The model prediction module obtains the growth temperature and humidity, disease index and number of larvae to be tested, and obtains the corresponding manure treatment volume prediction distribution through the manure treatment volume prediction model. The treatment volume with the highest probability of exceeding the preset threshold is selected as the manure treatment volume corresponding to the test data. The model building module specifically includes: The joint learning module models the features of black soldier fly frass treatment data, and selects the larval growth temperature and humidity, disease index and larval number that affect the black soldier fly frass treatment amount for temporal joint learning; The prediction distribution module stores the temporal information from multiple iterations in the final hidden layer state of the network. The data is passed as input to the output layer, where the probability distribution of sewage treatment volume is estimated using the softmax function. The formula is as follows: ； in, W , b These are the weights and biases.

2. The black soldier fly larvae treatment system based on intelligent control according to claim 1, characterized in that, The model training module specifically includes: The initialization module performs end-to-end training of the deep convolutional neural network, pre-training the model using image data and initializing it. The vector value computation module trains the temporal information storage network by performing forward computation. It uses a pre-trained deep convolutional neural network to extract features from image data obtained at different time periods and sequentially computes the output value of each neuron through the temporal information storage network. , , , , The values ​​of the five vectors; The weight gradient calculation module simultaneously fine-tunes all parameters of the deep convolutional neural network and the temporal information storage network, and calculates the error term δ value of each neuron in reverse. The backpropagation of the error term of the temporal information storage network includes two directions: one is backpropagation along time, starting from the current time t, to calculate the error term at each time step; the other is to propagate the error term to the upper layer of the feature extraction network, and calculate the gradient of each weight based on the corresponding error term.

3. The black soldier fly larvae treatment system based on intelligent control according to claim 1, characterized in that, The joint learning module specifically includes: The attribute settings module allows you to set time-invariant environmental attribute elements. K , L These represent the black soldier fly species and the type of excrement; a time-domain variable attribute element is defined, with temperature varying over time from t=1 to t=T. ,humidity and disease index ; The rules for updating the time-series information storage network are set in the rule setting module as follows: in, Let {K, L} be time-invariant environmental attribute elements. For attribute elements that vary in the time domain { , , }, and The transformation matrices for time-invariant and time-varying attributes are respectively given. The feature set of time-invariant environmental attribute factors for the time-series information storage network is as follows: ; The forward computation module normalizes the feature data, controlling the data dimension to between 0 and 1. It sets the number of layers in the time-series information storage network to correspond to specific time points, with each layer having both inputs and outputs. The first layer of the time-series information storage network takes time-invariant environmental attribute factors as input. The effects of this process are continuously transmitted to the prediction at each moment. After the image is processed by the deep feature extraction network, it is fused with the temporal change environmental attribute factors and used as the input of the n+1th layer of the temporal information storage network. The layers of the temporal information storage network are then input sequentially according to the order of shooting at different times, and the forward computation of the temporal information storage network is set.

4. The black soldier fly dung treatment system based on intelligent control according to claim 3, characterized in that, The forward calculation setting module specifically includes: Long-term storage unit Through the forgetting unit Remove forgotten information and set up forgetting units. Subject to external input at the current moment Short-time storage output of the previous time step Long-term storage of the previous time step The control is expressed as follows: ； , , , These represent the external input weights, short-term storage weights, long-term storage weights, and forgetting unit bias, respectively. External input at the current moment and the short-time storage output of the previous time step Calculate the new information at the current moment ; ； From the input unit Control will use the new information at the current moment Write to a long-term storage unit to create a new long-term storage unit. Its expression is as follows: ； ； in, by , , Control , , , These are, respectively, the input unit, long-term storage unit, forgetting unit, and output unit on the time-series information storage network; among them, The activation function represents the sigmoid shape. Indicates component multiplication. It is a weight matrix connecting different units; Activate long-term storage unit By output unit Control the storage that has been accumulated so far Select a portion of the relevant storage to generate the storage focus at this moment. Then output this part of the storage. , , ; Among them, the output unit by , and long-term storage at the current moment Control.

Images