Search

CN-121974731-A - AI-assisted kitchen waste fermentation process accurate control method

CN121974731ACN 121974731 ACN121974731 ACN 121974731ACN-121974731-A

Abstract

The application relates to an AI-assisted kitchen waste fermentation process accurate control method and system, wherein the method comprises the steps of collecting multisource sensing data in a kitchen waste fermentation reactor in real time, and synchronously acquiring environmental weather information and material initial physicochemical properties; the method comprises the steps of extracting dynamic change characteristics from multi-source sensing data, generating a fusion state vector by combining environmental weather information and initial physical and chemical properties of materials, inputting the fusion state vector into a pre-constructed fermentation process deduction model, obtaining a metabolic activity evaluation result of a current fermentation stage and a degree mark deviating from a reference track, generating a cooperative regulation instruction set aiming at ventilation intensity, pile turning frequency and water supplementing strategy based on the evaluation result and the mark, outputting the cooperative regulation instruction set to an execution mechanism of a fermentation reactor, and generating a fermentation process accurate control completion signal. The scheme can realize the accurate control of the kitchen waste fermentation process and improve the stability of the quality of fermentation products.

Inventors

  • ZHANG YANHUI
  • LI XIANG
  • Chang Xingyao

Assignees

  • 北京朝阳环境集团有限公司

Dates

Publication Date
20260505
Application Date
20251226

Claims (10)

  1. 1. An AI-assisted kitchen waste fermentation process precise control method is characterized by comprising the following steps: Acquiring multisource sensing data in the kitchen waste fermentation reactor in real time, and synchronously acquiring environmental meteorological information and initial physical and chemical properties of materials, wherein the multisource sensing data comprises a temperature time sequence, a humidity fluctuation curve, gas component concentration and stirring resistance signals; extracting dynamic change characteristics from the multi-source sensing data, and generating a fusion state vector by combining environmental weather information and initial physical and chemical properties of materials, wherein the fusion state vector comprises a temperature gradient change rate, humidity hysteresis deviation, a methane-carbon dioxide ratio and a material viscosity index; inputting the fusion state vector into a pre-constructed fermentation process deduction model to obtain a metabolism activity evaluation result of the current fermentation stage and a degree mark deviating from a reference track, wherein the metabolism activity evaluation result represents the activity level of a microbial community, the degree mark deviating from the reference track reflects the difference direction and amplitude of an actual working condition and an ideal fermentation path, and the fermentation process deduction model is a time sequence prediction model based on a long-short-period memory network; Generating a cooperative regulation instruction set aiming at ventilation intensity, pile turning frequency and water supplementing strategy based on the metabolic activity evaluation result and a degree mark deviating from a reference track, wherein the cooperative regulation instruction set comprises a relation between action priority ordering and time sequence coupling of each execution unit; Outputting the cooperative regulation instruction set to an executing mechanism of the fermentation reactor, and generating a fermentation process accurate control completion signal, wherein the fermentation process accurate control completion signal directly corresponds to a judging result of the quality stability of the kitchen waste fermentation product.
  2. 2. The method of claim 1, wherein the real-time collection of multi-source sensing data in the kitchen waste fermentation reactor and the simultaneous acquisition of environmental weather information and material initial physicochemical properties comprises: Carrying out space layered sampling on the interior of the fermentation reactor through a deployed distributed sensor array, obtaining temperature and humidity continuous records at different depth positions, and aligning all the records according to time stamps to form a three-dimensional space-time matrix, wherein the three-dimensional space-time matrix takes a height level as a first dimension, time as a second dimension and physical magnitude as a third dimension; Detecting the component proportion of the gas discharged from the top of the fermentation cavity through an infrared spectrum analysis device, extracting a volume fraction sequence of methane, carbon dioxide and hydrogen sulfide, and performing sliding window matching on the sequence and a standard fermentation gas spectrogram to generate a gas metabolism characteristic mark flow, wherein the gas metabolism characteristic mark flow indicates a switching node of an anaerobic or aerobic leading stage, and the standard fermentation gas spectrogram is defined based on a kitchen waste fermentation stage; Monitoring a current fluctuation curve of a driving motor during each pile turning operation, calculating a torque instantaneous change amplitude and a period average load, and mapping the torque instantaneous change amplitude and the period average load into a material structure compactness index, wherein the material structure compactness index is used for representing a pile body porosity attenuation trend; The method comprises the steps of calling data of the atmospheric pressure, the ambient temperature and the relative humidity on the same day from an external weather service platform interface, carrying out correlation modeling on the data and the surface temperature difference of a shell of a fermentation reactor, and generating a heat exchange compensation factor sequence, wherein the heat exchange compensation factor sequence is used for correcting external interference items of internal temperature measurement values; And (3) inputting the source type, the initial water content value and the estimated carbon nitrogen ratio value of the kitchen waste in the feeding stage, encoding the information into a material identity tag and binding the material identity tag to the current fermentation batch, wherein the material identity tag is used as an initial boundary condition for deduction in a subsequent state.
  3. 3. The method of claim 1, wherein extracting dynamic change features from the multi-source sensor data and generating a fusion state vector in combination with environmental weather information and material initial physicochemical properties comprises: Carrying out first-order differential operation on the temperature time sequence, identifying the duration of a temperature rise rate inflection point and a platform period, comparing the time interval between adjacent inflection points with a theoretical fermentation period template to generate a temperature phase offset, wherein the temperature phase offset quantifies the time sequence dislocation degree of an actual temperature rise rhythm and a standard curve, and the theoretical fermentation period template is constructed based on the carbon nitrogen ratio and the water content in a material identity label and comprises typical temperature rise/temperature reduction curve time sequence parameters; performing cross-correlation analysis on a humidity fluctuation curve and contemporaneous temperature change, calculating a humidity response delay time window, comparing the width of the time window with a preset threshold value interval, and generating a humidity hysteresis deviation sign, wherein the humidity hysteresis deviation sign indicates an unbalance direction between water evaporation and microorganism water consumption; Calculating the ratio of methane to carbon dioxide according to the volume fraction of the methane and the carbon dioxide in the gas component concentration, and tracking the monotonicity change trend of the ratio, and marking the ratio as a gas production peak period when the ratio continuously rises above a set slope, wherein the ratio of methane to carbon dioxide is used as a core criterion of anaerobic metabolism intensity; converting the stirring resistance signal into an energy consumption index of unit volume, and carrying out normalization treatment by combining the initial water content of the material to generate a material viscosity index, wherein the material viscosity index reflects the deterioration degree of the flowability of the pile; vector splicing is carried out on the temperature phase offset, the humidity hysteresis deviation sign, the methane-carbon dioxide ratio and the material viscosity index, and a carbon-nitrogen ratio estimated value in a material identity label is added to form a high-dimensional fusion state vector.
  4. 4. The method according to claim 1, wherein the inputting the fusion state vector into a pre-constructed fermentation process deduction model to obtain the metabolic activity assessment result of the current fermentation stage and the degree identifier of deviation from a reference track comprises: Establishing a multi-stage state transfer map in a pre-constructed fermentation process deduction model, wherein each node corresponds to a parameter combination range under a specific metabolic mode, projecting a fusion state vector to a nearest neighbor node of the map, and determining a current fermentation sub-stage, wherein the fermentation sub-stage comprises a starting period, a high temperature period, a decomposition period and a stabilization period; Calculating the mahalanobis distance between the fusion state vector and the current sub-stage center vector, and triggering a track deviation alarm if the distance exceeds a dynamic tolerance ellipsoid boundary, wherein the dynamic tolerance ellipsoid boundary adaptively contracts along with the fermentation time length, the dynamic tolerance ellipsoid is a multi-axis ellipsoid, and the tolerance radiuses of all axes are independently set according to the feature sensitivity; Based on the direction component of the track deviation alarm, specific cause types of temperature overshoot, low humidity or gas abnormality are decomposed, and a deviation reason code is generated, wherein the deviation reason code is used for guiding the selection of a subsequent regulation strategy; According to the theoretical metabolism rate curve of the current sub-stage, reversely deducing the difference between the theoretical heat production power of the microbial community and the actual measured value, normalizing the difference to obtain a metabolism activity assessment result, wherein the metabolism activity assessment result is output in a dimensionless exponential form; And carrying out weighted summation mapping on the metabolic activity evaluation result and the deviation reason code to generate a comprehensive state identifier, wherein the comprehensive state identifier uniquely corresponds to one regulation and control intervention mode.
  5. 5. The method of claim 1, wherein generating a coordinated regulation instruction set for ventilation intensity, pile-up-turning frequency and water replenishment strategy based on the metabolic activity assessment result and the degree of deviation from a reference trajectory comprises: when the current Xie Huoyue-degree evaluation result is lower than a lower limit threshold value and the deviation reason code indicates that the humidity is too low, preferentially activating a water supplementing executing mechanism and inhibiting ventilation action, and generating a water supplementing priority instruction, wherein the water supplementing priority instruction comprises spraying duration time and flow level; When the deviation reason code indicates that the temperature overshoots and the metabolism activity is in a high position, the ventilation intensity level is improved, the stack turning interval is shortened, and a heat dissipation strengthening instruction is generated, wherein the heat dissipation strengthening instruction prescribes the rotating speed gear of a fan and the starting period of a stack turning mechanical arm; constructing a conflict resolution rule base of three actions of ventilation, pile turning and water supplementing, and arbitrating according to the emergency degree of deviation reason coding when a plurality of instructions are triggered simultaneously to generate a conflict-free instruction sequence, wherein the conflict-free instruction sequence is used for defining the starting time and the continuous window of each action; Performing time sequence overlapping analysis on adjacent actions in the conflict-free instruction sequence, inserting a buffer waiting section if execution resource competition exists, and generating a final cooperative regulation instruction set, wherein the buffer waiting section ensures that the next action is started after the reset of the executing mechanism is completed; And adding a digital signature to the collaborative regulation instruction set and binding the fermentation batch identification to form a traceable regulation log, wherein the traceable regulation log is used for repeated disc and model iteration in the subsequent process.
  6. 6. A method according to claim 3, wherein the calculating the ratio of methane to carbon dioxide based on the volume fractions of the gas components and tracking the monotonicity trend thereof, and the marking of the gas production peak when the ratio continuously rises above a set slope, comprises: Performing moving average filtering on the methane volume fraction sequence, eliminating instantaneous measurement noise, obtaining a smooth methane concentration curve, performing point-by-point registration on the curve and an original carbon dioxide concentration curve, and generating a synchronous gas concentration pair; calculating the ratio of each pair of synchronous gas concentrations to form a methane-carbon dioxide ratio sequence, performing second-order differential operation on the sequence, and identifying the initial index of a ratio increase acceleration section, wherein the initial index corresponds to a gas production rate inflection point; intercepting a subsequent subsection of the ratio sequence from the initial index, fitting the linear regression slope of the subsection, and carrying out probability matching on the slope value and the reference slope distribution of the historical gas production peak period to generate a slope significance score; When the slope saliency score exceeds a preset confidence level, marking the current time point as a gas production peak period inlet, and generating a peak period inlet timestamp, wherein the peak period inlet timestamp is used for triggering an anaerobic maintenance strategy; and continuously monitoring the ratio sequence until the slope changes to a negative value, recording the intersection point of the positive rotation and the negative rotation of the slope as a peak period outlet, and generating the peak period duration, wherein the peak period duration is used as one of the judgment bases for starting the decomposition stage.
  7. 7. The method of claim 4, wherein calculating the mahalanobis distance between the fusion state vector and the current sub-phase center vector, and triggering a trajectory departure alert if the distance exceeds a dynamic tolerance ellipsoid boundary, wherein the dynamic tolerance ellipsoid boundary adaptively contracts with fermentation time, comprises: Extracting fusion state vector sample sets of each sub-stage from a historical successful fermentation batch, calculating a covariance matrix of each sample set, and defining a metric scale of the mahalanobis distance by using the covariance matrix, wherein the metric scale reflects a correlation structure of each characteristic dimension; the current fusion state vector and the current sub-stage center vector are subjected to difference to obtain a deviation vector, the inverse matrix of the covariance matrix is multiplied by the deviation vector, and then the inverse matrix is subjected to inner product with the inverse matrix, and a Markov distance value is obtained after square opening; Inquiring a preset tolerance decay function according to the fermentation run time, acquiring a tolerance radius corresponding to the current moment, and taking the tolerance radius as the length of a main shaft of a dynamic tolerance ellipsoid, wherein the tolerance decay function is in an exponential decline form; Comparing the Marshall distance value with the tolerance radius, and if the Marshall distance value is larger than the tolerance radius, judging that the track deviates, and generating a track deviation Boolean mark, wherein the track deviation Boolean mark is used for activating a regulation and control decision module; Recording the ratio of the mahalanobis distance value and the tolerance radius when each track deviation occurs, and storing the ratio into a process anomaly database, wherein the process anomaly database is used for optimizing tolerance decay function parameters of subsequent batches.
  8. 8. The method of claim 5, wherein when the deviation cause code indicates a temperature overshoot and the metabolic activity is high, increasing the ventilation intensity level and shortening the stack turning interval, generating a heat dissipation enhancement command, comprising: analyzing a temperature overshoot identification bit in the deviation reason code, confirming the duration of the temperature measured value exceeding the upper limit threshold of the current sub-stage, and mapping the duration into a ventilation intensity increment grade, wherein the ventilation intensity increment grade determines the fan rotating speed lifting amplitude; Reading a numerical value of a metabolic activity evaluation result, and if the numerical value is higher than a high-order threshold value, starting a forced heat dissipation mode to generate a fan full-power operation signal, wherein the forced heat dissipation mode covers a conventional temperature control logic; judging the air permeability state of the pile body according to the material viscosity index, if the air permeability is deteriorated, synchronously triggering pile turning action, and calculating the pile turning interval shortening proportion, wherein the pile turning interval shortening proportion is positively related to the viscosity index; the method comprises the steps of performing time alignment on a rotating speed gear of a fan and a starting period of a turning mechanical arm, ensuring that the maximum air volume is started to purge immediately after turning is completed, and generating a time sequence synchronous control word, wherein the time sequence synchronous control word comprises a delay counter initial value from the end of turning to the start of the fan; and packaging the heat dissipation strengthening instruction into a control message with an aging period, and setting the failure time of the message as the end point of the high-temperature continuous prediction window, wherein the aging period prevents metabolic interruption caused by excessive cooling.
  9. 9. The method of claim 6, wherein intercepting subsequent subsections of the ratio sequence from the start index, fitting a linear regression slope of the subsections, probability matching the slope value to a baseline slope distribution of historical gas production peak periods, generating a slope saliency score, comprising: The method comprises the steps of calling a historical gas production peak period slope sample set of similar materials in the same sub-stage from a process database, calculating an experience cumulative distribution function of the sample set, and substituting a current slope value into the function to obtain a cumulative probability value; Comparing the cumulative probability value with the upper limit and the lower limit of the bilateral confidence interval, judging that the gas production is abnormal and high if the cumulative probability value exceeds the upper limit, judging that the gas production is insufficient if the cumulative probability value is lower than the lower limit, and generating a slope abnormal type identifier, wherein the slope abnormal type identifier is used for distinguishing the metabolic overstress and the inhibition state; Calculating a deviation metric based on the cumulative probability value and the distance of the confidence interval, compressing the deviation metric to a zero-to-one interval through an S-shaped function, and generating a slope saliency score, wherein the closer the slope saliency score is to the one, the more typical the slope saliency score is; when the slope saliency score is lower than a preset lower limit, a ratio sequence monitoring window is prolonged and the slope is re-fitted, so that a slope re-estimation trigger signal is generated, wherein the slope re-estimation trigger signal avoids misjudging the peak time due to short-term fluctuation; And binding and storing the slope saliency score and the peak period entry timestamp to form a gas production characteristic file, wherein the gas production characteristic file is used for updating the reference slope distribution of similar materials.
  10. 10. An AI-assisted kitchen waste fermentation process accurate control system, characterized in that the system comprises: The acquisition module is used for acquiring multi-source sensing data in the kitchen waste fermentation reactor in real time and synchronously acquiring environmental weather information and initial physical and chemical properties of materials, wherein the multi-source sensing data comprises a temperature time sequence, a humidity fluctuation curve, gas component concentration and stirring resistance signals; The characteristic processing module is used for extracting dynamic change characteristics from the multi-source sensing data and generating a fusion state vector by combining environmental weather information and initial physical and chemical properties of materials, wherein the fusion state vector comprises a temperature gradient change rate, humidity hysteresis deviation, a methane-carbon dioxide ratio and a material viscosity index; The deduction processing module is used for inputting the fusion state vector into a pre-constructed fermentation process deduction model to obtain a metabolism activity evaluation result of the current fermentation stage and a degree mark deviating from a reference track, wherein the metabolism activity evaluation result represents the activity level of a microbial community, the degree mark deviating from the reference track reflects the difference direction and amplitude of an actual working condition and an ideal fermentation path, and the fermentation process deduction model is a time sequence prediction model based on a long-short-period memory network; The instruction generation module is used for generating a cooperative regulation instruction set aiming at ventilation intensity, pile turning frequency and water supplementing strategy based on the metabolic activity evaluation result and the degree mark deviating from the reference track, wherein the cooperative regulation instruction set comprises a relation between action priority ordering and time sequence coupling of each execution unit; And the sending module is used for outputting the cooperative regulation instruction set to an executing mechanism of the fermentation reactor and generating a fermentation process accurate control completion signal, wherein the fermentation process accurate control completion signal directly corresponds to a judging result of the quality stability of the kitchen waste fermentation product.

Description

AI-assisted kitchen waste fermentation process accurate control method Technical Field The application relates to the technical field of environmental protection and information, in particular to an AI-assisted kitchen waste fermentation process accurate control method and system. Background In the field of kitchen waste treatment, fermentation technology is an important way for realizing the recycling utilization of kitchen waste. Through the fermentation process, kitchen garbage can be converted into organic fertilizer, biogas and other useful resources. At present, the traditional kitchen waste fermentation process mainly depends on manual experience and a simple timing control mode. The operator adjusts the ventilation, turning, etc. operations according to the general time and some basic sensory judgment, such as temperature, smell, etc. However, this conventional approach has a number of problems. Because of subjectivity and limitation of manual judgment, it is difficult to accurately grasp the real-time state of the fermentation process. The differences in experience of the different operators can also lead to instability of the fermentation effect. To solve these problems, some techniques have attempted to introduce sensors to monitor some parameters of the fermentation process, such as temperature and humidity, and to make simple control according to preset fixed thresholds. However, these solutions still have a critical problem that the complex dynamic changes of the fermentation process cannot be reflected comprehensively and accurately. The fermentation process is affected by a combination of factors including gas composition, material properties, and the like in addition to temperature and humidity. The control is carried out only by means of limited parameters and fixed threshold values, the continuous change of actual working conditions in the fermentation process is difficult to adapt, the quality of the fermentation product is unstable, and the recycling potential of kitchen waste cannot be fully exerted. Disclosure of Invention The application mainly aims to provide an AI-assisted kitchen waste fermentation process accurate control method and system, which can comprehensively collect multi-source information in the fermentation process, accurately analyze fermentation states, realize cooperative regulation and control of operations such as ventilation, turning and water supplementing, and improve the quality stability of kitchen waste fermentation products. In order to achieve the above purpose, the embodiment of the invention provides an AI-assisted kitchen waste fermentation process accurate control method, which comprises the following steps: Acquiring multisource sensing data in the kitchen waste fermentation reactor in real time, and synchronously acquiring environmental meteorological information and initial physical and chemical properties of materials, wherein the multisource sensing data comprises a temperature time sequence, a humidity fluctuation curve, gas component concentration and stirring resistance signals; extracting dynamic change characteristics from the multi-source sensing data, and generating a fusion state vector by combining environmental weather information and initial physical and chemical properties of materials, wherein the fusion state vector comprises a temperature gradient change rate, humidity hysteresis deviation, a methane-carbon dioxide ratio and a material viscosity index; inputting the fusion state vector into a pre-constructed fermentation process deduction model to obtain a metabolism activity evaluation result of the current fermentation stage and a degree mark deviating from a reference track, wherein the metabolism activity evaluation result represents the activity level of a microbial community, the degree mark deviating from the reference track reflects the difference direction and amplitude of an actual working condition and an ideal fermentation path, and the fermentation process deduction model is a time sequence prediction model based on a long-short-period memory network; Generating a cooperative regulation instruction set aiming at ventilation intensity, pile turning frequency and water supplementing strategy based on the metabolic activity evaluation result and a degree mark deviating from a reference track, wherein the cooperative regulation instruction set comprises a relation between action priority ordering and time sequence coupling of each execution unit; Outputting the cooperative regulation instruction set to an executing mechanism of the fermentation reactor, and generating a fermentation process accurate control completion signal, wherein the fermentation process accurate control completion signal directly corresponds to a judging result of the quality stability of the kitchen waste fermentation product. In summary, by adopting the technical scheme of the application, the actual condition of the fermentation process can be compr