Intelligent control method of coal caving hydraulic support

Abstract

The present application relates to a kind of intelligent control methods of coal hydraulic support, belong to coal mine intelligent technology field.Method includes: obtaining the original audio signal of current state of fully mechanized caving face and after pretreatment, the power spectrum of each frequency channel of each frame signal is calculated;Dynamic tilt parameter of each frequency channel of each frame signal is calculated, and the time-domain impulse response of each frequency channel of each frame signal is calculated using dynamic compression filter bank;According to time-domain impulse response and its power spectrum, the compression energy value of each frequency channel of each frame signal is calculated;Spectrum chart is generated by gathering compression energy value, and the recognition probability distribution result of each frame signal is obtained by using improved residual shrinkage model processing;The control strategy of current state of coal hydraulic support is generated, and coal hydraulic support is controlled to execute control strategy.The present application processes audio signal based on dynamic compression filter bank, and the accuracy of identification result is improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent coal mining technology, and in particular to an intelligent control method for a coal discharge hydraulic support. Background Technology

[0002] Coal release, a unique process in fully mechanized longwall mining, is crucial for upgrading the technology to intelligent operation. In traditional longwall mining, the hydraulic supports for coal release are typically manual or electro-hydraulic controlled. Workers rely on experience, listening to the sounds of coal and gangue colliding to determine whether to close the release port or stop release. The harsh working environment, high dust levels, narrow field of vision, and complex process make over- or under-release common, hindering precise and intelligent control of the hydraulic supports.

[0003] Furthermore, the control strategy of the hydraulic support for coal discharge is based on the modal identification results of coal and gangue impact sound in the fully mechanized longwall face. However, the background noise intensity generated by the coal mining machine, scraper conveyor, and hydraulic support in the fully mechanized longwall face is high, masking the transient acoustic characteristics of coal and gangue impact sound, resulting in a low signal-to-noise ratio of the collected audio. In addition, the fully mechanized longwall face has a complex scenario of multi-source noise superposition and dynamic changes in coal discharge conditions, further leading to low accuracy of coal and gangue impact sound modal identification and poor noise resistance, making it impossible to provide an accurate control strategy for the hydraulic support for coal discharge. Summary of the Invention

[0004] To solve the above-mentioned technical problems, the present invention provides an intelligent control method for a coal discharge hydraulic support. The technical solution of the present invention is as follows: A method for intelligent control of a coal discharge hydraulic support, comprising: S1, acquire the raw audio signal of the current state of the fully mechanized amplification working face; S2, preprocess the original audio signal to obtain multi-frame signals, and calculate the power spectrum of each frequency channel of each frame signal; S3, calculate the dynamic tilt parameter of each frequency channel of each frame signal based on the power spectrum of each frequency channel of each frame signal, and calculate the time-domain impulse response of each frequency channel of each frame signal using a dynamically compressed filter bank based on the dynamic tilt parameter of each frequency channel of each frame signal. S4. Calculate the compressed energy value of each frequency channel of each frame signal based on the time-domain impulse response and power spectrum of each frequency channel of each frame signal. S5, the compressed energy values ​​of each frequency channel of each frame signal are collected to generate a spectrogram, and the spectrogram is processed by an improved residual shrinkage model to obtain the recognition probability distribution results of each frame signal; S6 generates a control strategy for the current state of the coal discharge hydraulic support based on the recognition probability distribution of each frame signal, and controls the coal discharge hydraulic support to execute the control strategy.

[0005] Preferably, S2 includes: S21, the original audio signal is pre-emphasized using a transfer function to obtain the pre-emphasized audio signal; S22, the pre-emphasized audio signal is split according to the preset length to obtain multi-frame unit audio signals, and each frame unit audio signal is processed by a window function to obtain each frame signal; S23 uses a dynamically compressed filter bank to calculate the power spectrum of each frequency channel of the signal in each frame.

[0006] Preferably, step S23 calculates the power spectrum of the j-th frequency channel of the m-th frame signal. When, this is achieved through formulas (1) to (3): (1); (2); (3); In formulas (1) to (3), Indicates frequency point index, Let j represent the actual frequency of the k-th frequency point of the m-th frame signal, j represent the frequency channel index, n represent the n-th time node, and N represent the total number of time nodes of the m-th frame signal. This represents the signal value corresponding to the nth time node in the m-th frame of the signal, where i represents the imaginary unit. This represents the power response weight at the k-th frequency point on the j-th frequency channel of the m-th frame signal. Let represent the center frequency of the j-th filter in the dynamically compressed filter bank. This represents the equivalent rectangular bandwidth of the j-th filter. This represents the minimum tilt value of the dynamically compressed filter bank obtained by fitting the filter function. It is the arctangent function. This represents an exponential function with the natural constant e as its base.

[0007] Preferably, S3 includes: S31, calculate the dynamic tilt parameter of each frequency channel of each frame signal based on the power spectrum of each frequency channel of each frame signal; S32 inputs the dynamic tilt parameters of each frequency channel of each frame signal into the dynamically compressed filter bank to obtain the time-domain impulse response of each frequency channel of each frame signal.

[0008] Preferably, step S31 is based on the power spectrum of the j-th frequency channel of the m-th frame signal. Calculate the dynamic tilt parameter of the j-th frequency channel of the m-th frame signal. When, this is achieved through formula (4): (4); In formula (4), This represents the minimum tilt value of the dynamically compressed filter bank obtained by fitting the filter function. This represents the magnitude of the dynamic tilt parameter variation range, where S represents the tilt parameter of the dynamically compressed filter bank, and O represents the offset parameter of the dynamically compressed filter bank. This represents an exponential function with the natural constant e as its base.

[0009] Preferably, step S32 involves dynamically adjusting the tilt parameter of the j-th frequency channel of the m-th frame signal. The time-domain impulse response of the j-th frequency channel of the m-th frame signal is obtained by inputting it into the filter bank of the dynamic compression. This is achieved through formula (5): (5); In formula (5), Let represent the center frequency of the j-th filter in the dynamically compressed filter bank. Let represent the equivalent rectangular bandwidth of the j-th filter, A represent the preset amplitude of the dynamically compressed filter bank, t represent time, q represent the order of the dynamically compressed filter bank, and b represent the preset attenuation factor. This represents the initial phase, and i represents the imaginary unit. Represents the logarithmic function. This represents an exponential function with the natural constant e as its base.

[0010] Preferably, S4 includes: The time-domain impulse response and its power spectrum of each frequency channel of each frame signal are weighted and summed in the frequency domain by using a frequency domain multiplication operation to obtain the compressed energy value of each frequency channel of each frame signal.

[0011] Preferably, the improved residual shrinkage model includes an input layer, a convolutional layer, multiple stacked residual shrinkage units, a global average pooling layer, and a fully connected classification layer. Each residual shrinkage unit includes a global average pooling unit, a fully connected unit, an activation unit, and an output unit. S5 includes: S51, calculate the Wiener entropy value of each frequency channel based on the compressed energy value of each frequency channel of all frame signals; S52, the spectrogram is input from the input layer and the feature is extracted through the convolutional layer to obtain the input features of each frame of signal; S53, through a multi-layer stacked residual shrinking unit, performs adaptive denoising on the input features of each frame signal in sequence, and the last residual shrinking unit outputs the denoised features of each frame signal. S54 converts the denoising features of each frame signal into feature vectors of each frame signal through a global average pooling layer, and outputs the recognition probability distribution of each frame signal by a fully connected classification layer. In this process, the stacked residual shrinking unit performs adaptive denoising on the input features of each frame of signal, while the global average pooling unit performs global average pooling on the input features of each frame of signal to obtain the pooling value of each frequency channel. Based on the pooling value of each frequency channel and the Wiener entropy value of each frequency channel, an enhancement vector for each frequency channel is constructed. The fully connected unit converts the enhancement vector of each frequency channel into a customized weight for each frequency channel. The activation unit calculates the frequency channel threshold based on the customized weight of each frequency channel and the pooling value of each frequency channel. The output unit performs adaptive denoising on the input features of each frame of signal using the frequency channel threshold.

[0012] Preferably, S51 includes: S511, calculate the arithmetic mean of the compressed energy values ​​of each frequency channel based on the sum of the compressed energy values ​​of each frequency channel of all frame signals; S512, calculate the geometric mean of the compressed energy value of each frequency channel based on the sum of the logarithms of the compressed energy values ​​of each frequency channel of all frame signals; S513, calculate the ratio of the geometric mean to the arithmetic mean of the compressed energy values ​​for each frequency channel to obtain the Wiener entropy value for each frequency channel.

[0013] Preferably, S6 includes: S61, determine whether the probability of coal release in the recognition probability distribution result of any frame signal is greater than the preset threshold. If the probability of coal release in the recognition probability distribution result of any frame signal is greater than the preset threshold and the probability of gangue release in the recognition probability distribution results of multiple consecutive frames of signals is less than the preset threshold, then determine that the current state of the fully mechanized mining face is coal release. S62, determine whether the probability of waste rock release in the recognition probability distribution results of consecutive multi-frame signals is greater than a preset threshold. If the probability of waste rock release in the recognition probability distribution results of consecutive multi-frame signals is greater than the preset threshold, then determine that the current state of the fully mechanized mining face is waste rock release. S63 generates a control strategy for the current state of the coal release hydraulic support based on the current state of the fully mechanized longwall face, and controls the coal release hydraulic support to execute the control strategy.

[0014] All of the above-mentioned optional technical solutions can be combined arbitrarily, and the present invention will not provide a detailed description of the structure after each combination.

[0015] By means of the above solution, the beneficial effects of the present invention are as follows: This invention provides an intelligent control method for coal release hydraulic supports. By preprocessing the raw audio signal of the current state of the fully mechanized longwall face, the power spectrum of each frequency channel in each frame is obtained. The dynamic tilt parameters of each frequency channel in each frame are calculated, and the time-domain impulse response and compressed energy value of each frequency channel in each frame are calculated using a dynamically compressed filter bank. An improved residual contraction model is then used to process the spectrum, resulting in the recognition probability distribution of each frame. Based on this distribution, a control strategy for the current state of the coal release hydraulic support is generated, providing an intelligent control method for the coal release hydraulic support. Compared with traditional methods, this invention identifies the current coal release condition based on the recognition probability distribution and generates a control strategy for the coal release hydraulic support accordingly. This achieves full automation and intelligence of the coal release process, improving the recovery rate and coal quality of the fully mechanized longwall face. The use of a dynamically compressed filter bank, inspired by the human ear, reduces the impact of background noise at the fully mechanized longwall face, improving the accuracy of the recognition results.

[0016] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0017] Figure 1 This is a flowchart of an intelligent control method for a coal discharge hydraulic support provided by the present invention. Detailed Implementation

[0018] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0019] like Figure 1 As shown in the figure, an intelligent control method for a coal discharge hydraulic support provided in this embodiment of the invention includes steps S1 to S6: S1, acquire the raw audio signal of the current state of the fully mechanized amplification working face.

[0020] Specifically, the raw audio signal of the current state of the fully mechanized mining face is acquired using an audio sensor. The audio sensor can be configured near the scraper conveyor. The raw audio signal of the current state is the raw audio signal acquired in the current acquisition cycle. The audio sensor acquires the raw audio signal at a fixed acquisition cycle. The specific duration of the acquisition cycle can be set as needed.

[0021] S2 preprocesses the original audio signal to obtain multiple frames of signal and calculates the power spectrum of each frequency channel of each frame of signal.

[0022] Specifically, in this embodiment of the invention, preprocessing is achieved through frame division and a filter bank of dynamic compression. One frame of signal contains signal values ​​corresponding to multiple time nodes.

[0023] In one specific embodiment, S2 includes: S21, the original audio signal is pre-emphasized using a transfer function to obtain the pre-emphasized audio signal.

[0024] Specifically, the signal value of the original audio signal at the g-th time node is obtained through the transfer function. Perform pre-emphasis processing to obtain the signal value of the pre-emphasized audio signal at the g-th time node. The calculation formula is: ;in, This represents the signal value of the original audio signal at the (g-1)th time node; This represents the pre-weighting coefficient, which is generally 0.9-1.0.

[0025] Pre-emphasis processing can suppress low-frequency background noise in the fully mechanized mining face equipment and compensate for high-frequency attenuation and enhance high-frequency signals.

[0026] S22, the pre-emphasized audio signal is split according to the preset length to obtain multi-frame unit audio signals, and each frame unit audio signal is processed by a window function to obtain each frame signal.

[0027] Specifically, the preset length is the pre-defined time interval between frames, typically 20ms. To reduce leakage of the audio signal per frame, a window function can be applied to each frame to eliminate discontinuities between the start and end points of the audio signal per frame. This is achieved through the window function. Processing the Frame unit audio signal , obtained the Frame signal The calculation formula is: .

[0028] By splitting the pre-emphasized signal into individual frames, the requirements for subsequent spectrum analysis can be met.

[0029] S23 uses a dynamically compressed filter bank to calculate the power spectrum of each frequency channel of the signal in each frame.

[0030] Specifically, S23 calculates the power spectrum of the j-th frequency channel of the m-th frame signal. When, this is achieved through formulas (1) to (3): (1); (2); (3); In formulas (1) to (3), Indicates frequency point index, Let j represent the actual frequency of the k-th frequency point of the m-th frame signal, j represent the frequency channel index, n represent the n-th time node, and N represent the total number of time nodes of the m-th frame signal. This represents the signal value corresponding to the nth time node in the m-th frame of the signal, where i represents the imaginary unit. This represents the power response weight at the k-th frequency point on the j-th frequency channel of the m-th frame signal. Let represent the center frequency of the j-th filter in the dynamically compressed filter bank. This represents the equivalent rectangular bandwidth of the j-th filter. This represents the minimum tilt value of the dynamically compressed filter bank obtained by fitting the filter function. It is the arctangent function. This represents an exponential function with the natural constant e as its base.

[0031] S3, calculate the dynamic tilt parameter of each frequency channel of each frame signal based on the power spectrum of each frequency channel of each frame signal, and calculate the time-domain impulse response of each frequency channel of each frame signal using a dynamically compressed filter bank based on the dynamic tilt parameter of each frequency channel of each frame signal.

[0032] Specifically, the dynamic tilt parameter represents the degree of tilt of the power spectrum of each frequency channel of the signal in each frame as the frequency channel changes. In this embodiment of the invention, the dynamically compressed filter bank employs a Gammachirp filter bank. The Gammachirp filter bank is an auditory filter bank based on the characteristics of human auditory perception, capable of mimicking the nonlinear and dynamic characteristics of human hearing. By employing a Gammachirp filter bank, the robustness of coal gangue identification in high-noise backgrounds can be improved.

[0033] In one specific embodiment, S3 includes: S31, calculate the dynamic tilt parameter of each frequency channel of each frame signal based on the power spectrum of each frequency channel of each frame signal.

[0034] In one specific embodiment, S31 is based on the power spectrum of the j-th frequency channel of the m-th frame signal. Calculate the dynamic tilt parameter of the j-th frequency channel of the m-th frame signal. When, this is achieved through formula (4): (4); In formula (4), This represents the minimum tilt value of the dynamically compressed filter bank obtained by fitting the filter function. This represents the magnitude of the dynamic tilt parameter variation range, where S represents the tilt parameter of the dynamically compressed filter bank, and O represents the offset parameter of the dynamically compressed filter bank. This represents an exponential function with the natural constant e as its base.

[0035] Specifically, when When the signal is very small, the j-th frequency channel of the m-th frame is a weak signal. Approximately equal to ;when When the signal is very large, the j-th frequency channel of the m-th frame signal is a strong signal. Approximately equal to and sum.

[0036] in addition, The corresponding bandwidth is the narrowest, which amplifies weak signals; The maximum tilt value corresponds to the widest bandwidth and acts as a compressor and saturator for strong signals. S and O determine the steepness of the dynamic tilt parameter. Specifically, when... When the value is greater than 0, the high-frequency side attenuation of the j-th filter corresponding to the j-th frequency channel is steeper.

[0037] S32 inputs the dynamic tilt parameters of each frequency channel of each frame signal into the dynamically compressed filter bank to obtain the time-domain impulse response of each frequency channel of each frame signal.

[0038] In one specific embodiment, S32 dynamically tilts the j-th frequency channel of the m-th frame signal. The time-domain impulse response of the j-th frequency channel of the m-th frame signal is obtained by inputting it into the filter bank of the dynamic compression. This is achieved through formula (5): (5); In formula (5), Let represent the center frequency of the j-th filter in the dynamically compressed filter bank. The equivalent rectangular bandwidth of the j-th filter is represented by the following formula: A represents the preset amplitude of the dynamically compressed filter bank, t represents time, q represents the order of the dynamically compressed filter bank, and b represents the preset attenuation factor. This represents the initial phase, and i represents the imaginary unit. Represents the logarithmic function. This represents an exponential function with the natural constant e as its base.

[0039] The embodiments of the present invention employ a signal processing method based on a dynamically compressed Gammachirp filter bank. By dynamically compressing to simulate the human ear, it can automatically enhance weak signals and suppress strong signals, thereby achieving adaptive adjustment of the Gammachirp filter bank based on a dynamic tilting mechanism according to the intensity of the input audio signal.

[0040] S4. Calculate the compressed energy value of each frequency channel of each frame signal based on the time-domain impulse response and power spectrum of each frequency channel of each frame signal.

[0041] In a specific embodiment, S4 includes: using a frequency domain multiplication operation to perform a weighted summation of the time-domain impulse response and its power spectrum of each frequency channel of each frame signal in the frequency domain to obtain the compressed energy value of each frequency channel of each frame signal.

[0042] It should be noted that, in addition to using frequency domain multiplication, embodiments of the present invention can also use convolution operations to perform weighted summation of the time-domain impulse response and power spectrum of each frequency channel of each frame signal in the frequency domain to obtain the compressed energy value of each frequency channel of each frame signal.

[0043] S5: The compressed energy values ​​of each frequency channel of each frame signal are collected to generate a spectrogram. An improved residual shrinkage model is used to process the spectrogram to obtain the recognition probability distribution of each frame signal.

[0044] Specifically, the spectrogram is a matrix whose number of rows corresponds to the number of filters in the Gammachirp filter bank, and its number of columns corresponds to the number of frames in the signal. The improved residual shrinkage model is a neural network model pre-trained from historical spectrograms.

[0045] In a specific embodiment, the improved residual shrinkage model includes an input layer, a convolutional layer, multiple stacked residual shrinkage units, a global average pooling layer, and a fully connected classification layer. Each residual shrinkage unit includes a global average pooling unit, a fully connected unit, an activation unit, and an output unit. S5 includes: S51, calculate the Wiener entropy value of each frequency channel based on the compressed energy value of each frequency channel of all frame signals.

[0046] In one specific embodiment, S51 includes: S511 calculates the arithmetic mean of the compressed energy values ​​of each frequency channel based on the sum of the compressed energy values ​​of each frequency channel of all frame signals.

[0047] Specifically, based on the sum of the compressed energy values ​​of the j-th frequency channel of all frame signals... Calculate the arithmetic mean of the compressed energy values ​​of the j-th frequency channel. When the time is right, the calculation formula is: Specifically, M represents the number of frames in the signal.

[0048] S512 calculates the geometric mean of the compressed energy value of each frequency channel based on the sum of the logarithms of the compressed energy values ​​of each frequency channel of all frame signals.

[0049] Specifically, it is based on the sum of the logarithms of the compressed energy values ​​of the j-th frequency channel of all frame signals. Calculate the geometric mean of the compressive energy values ​​of the j-th frequency channel. When the time is right, the calculation formula is: .

[0050] S513, calculate the ratio of the geometric mean to the arithmetic mean of the compressed energy values ​​for each frequency channel to obtain the Wiener entropy value for each frequency channel.

[0051] Based on the specific calculation formula above, the Wiener entropy value of the j-th frequency channel The calculation formula is: .

[0052] S52 inputs the spectrogram from the input layer and extracts features through the convolutional layer to obtain the input features of each frame of signal.

[0053] S53 uses a series of stacked residual shrinking units to perform adaptive denoising on the input features of each frame of signal, and the last residual shrinking unit outputs the denoised features of each frame of signal.

[0054] S54 converts the denoising features of each frame signal into feature vectors of each frame signal through a global average pooling layer, and outputs the recognition probability distribution of each frame signal by a fully connected classification layer. In this process, the stacked residual shrinking unit performs adaptive denoising on the input features of each frame of signal, while the global average pooling unit performs global average pooling on the input features of each frame of signal to obtain the pooling value of each frequency channel. Based on the pooling value of each frequency channel and the Wiener entropy value of each frequency channel, an enhancement vector for each frequency channel is constructed. The fully connected unit converts the enhancement vector of each frequency channel into a customized weight for each frequency channel. The activation unit calculates the frequency channel threshold based on the customized weight of each frequency channel and the pooling value of each frequency channel. The output unit performs adaptive denoising on the input features of each frame of signal using the frequency channel threshold.

[0055] Specifically, in conjunction with the above embodiments, a certain residual shrinkage unit, based on the pooling value of the j-th frequency channel... Wiener entropy value of the j-th frequency channel The enhancement vector constructed for the j-th frequency channel is .

[0056] When stacked residual shrinking units adaptively denoise the input features of a frame signal to obtain the denoised features of that frame signal, they use frequency channel thresholding. This is achieved by constructing a soft thresholding function, specifically as follows: ; in, This indicates the input characteristics of the signal in this frame. This represents the denoising characteristics of the signal in that frame. The soft thresholding function not only sets tiny input features (noise) close to zero to zero, but also preserves significant positive and negative input features.

[0057] The probability distribution of each frame of signal includes the probability of coal release and the probability of gangue release. For example, the probability distribution of a certain frame of signal is [P(coal release), P(gangue release)], where P(coal release) represents the probability of coal release in that frame of signal and P(gangue release) represents the probability of gangue release in that frame of signal.

[0058] Stacked residual shrinking units fuse Wiener entropy and pooling values ​​by constructing enhancement vectors. A soft thresholding function adaptively adjusts the frequency channel threshold based on the Wiener entropy of each channel, and layer-by-layer noise reduction is achieved through stacked residual shrinking units (adaptively adjusted via soft thresholding at each layer), enabling intelligent signal noise reduction. Improving the residual shrinking model with Wiener entropy enhances the ability to perform feature extraction and intelligent adaptive noise reduction based on auditory bionics, significantly improving the accuracy and robustness of coal gangue identification under strong noise conditions.

[0059] S6 generates a control strategy for the current state of the coal discharge hydraulic support based on the recognition probability distribution of each frame signal, and controls the coal discharge hydraulic support to execute the control strategy.

[0060] Specifically, the control strategies for the coal discharge hydraulic support include two types: coal discharge (start) and window closing (end). When gangue appears in the fully mechanized coal discharge face, the control strategy is to close the window.

[0061] In one specific embodiment, S6 includes: S61, determine whether the probability of coal release in the recognition probability distribution result of any frame signal is greater than a preset threshold. If the probability of coal release in the recognition probability distribution result of any frame signal is greater than the preset threshold and the probability of gangue release in the recognition probability distribution results of multiple consecutive frames of signals is less than the preset threshold, then determine that the current state of the fully mechanized mining face is coal release.

[0062] Specifically, when multiple consecutive frames of signals are identified as gangue discharge, it is considered that gangue has begun to mix into the fully mechanized longwall face in large quantities, and a window-closing control strategy needs to be generated immediately. Otherwise, the coal discharge state of the hydraulic support is maintained to ensure the accuracy and stability of the control strategy, providing the hydraulic support with an accurate coal discharge and window-closing control strategy. The preset threshold is generally 75%.

[0063] S62, determine whether the probability of waste rock release in the recognition probability distribution results of multiple consecutive frames of signals is greater than a preset threshold. If the probability of waste rock release in the recognition probability distribution results of multiple consecutive frames of signals is greater than the preset threshold, then determine that the current state of the fully mechanized mining face is waste rock release.

[0064] S63 generates a control strategy for the current state of the coal release hydraulic support based on the current state of the fully mechanized longwall face, and controls the coal release hydraulic support to execute the control strategy.

[0065] Specifically, after the control strategy is executed by the coal discharge hydraulic support, the complete operation of the coal discharge hydraulic support in discharging coal or closing the window is monitored, including initialization, circulating coal discharge, tail beam swing, waiting sequence, etc.

[0066] In summary, the intelligent control method for a coal discharge hydraulic support provided in this embodiment of the invention has the following beneficial effects: By dynamically compressing the Gammachirp filter bank to process signals, the active gain control and frequency sharpening mechanism of the human ear under different sound pressure levels can be dynamically simulated. It integrates nonlinear compression and adaptive frequency analysis, dynamically amplifying weak signals in coal / gangue disposal and other working conditions in the time and frequency domains, while suppressing strong background noise signals. The generated spectrogram can capture the subtle differences in timbre, resonant frequency and other properties of coal and gangue. It transforms the auditory experience of underground coal miners in "distinguishing objects by sound" into a series of quantifiable and optimizable dynamic parameters, providing higher quality and more separable inputs for the backend network.

[0067] This invention creatively combines Wiener entropy values ​​with an improved residual shrinkage model, enabling the output of the recognition probability distribution of each frame of signal based on the improved residual shrinkage model of Wiener entropy.

[0068] This invention also proposes a multi-mode, multi-cycle parameter configurable programmable control strategy. Based on the recognition probability distribution of each frame signal, it directly sends coal release / window closing instructions to the coal release hydraulic support, realizing full automation and intelligence of coal release on the fully mechanized coal release face, improving recovery rate and coal quality, and enhancing the reliability of coal release hydraulic support control.

[0069] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An intelligent control method for a coal discharge hydraulic support, characterized in that, include: S1, acquire the raw audio signal of the current state of the fully mechanized amplification working face; S2, preprocess the original audio signal to obtain multi-frame signals, and calculate the power spectrum of each frequency channel of each frame signal; S3, calculate the dynamic tilt parameter of each frequency channel of each frame signal based on the power spectrum of each frequency channel of each frame signal, and calculate the time-domain impulse response of each frequency channel of each frame signal using a dynamically compressed filter bank based on the dynamic tilt parameter of each frequency channel of each frame signal. S4. Calculate the compressed energy value of each frequency channel of each frame signal based on the time-domain impulse response and power spectrum of each frequency channel of each frame signal. S5, the compressed energy values ​​of each frequency channel of each frame signal are collected to generate a spectrogram, and the spectrogram is processed by an improved residual shrinkage model to obtain the recognition probability distribution results of each frame signal; S6 generates a control strategy for the current state of the coal discharge hydraulic support based on the recognition probability distribution of each frame signal, and controls the coal discharge hydraulic support to execute the control strategy.

2. The intelligent control method for a coal discharge hydraulic support according to claim 1, characterized in that, S2 includes: S21, the original audio signal is pre-emphasized using a transfer function to obtain the pre-emphasized audio signal; S22, the pre-emphasized audio signal is split according to the preset length to obtain multi-frame unit audio signals, and each frame unit audio signal is processed by a window function to obtain each frame signal; S23 uses a dynamically compressed filter bank to calculate the power spectrum of each frequency channel of the signal in each frame.

3. The intelligent control method for a coal discharge hydraulic support according to claim 2, characterized in that, S23 calculates the power spectrum of the j-th frequency channel of the m-th frame signal. When, this is achieved through formulas (1) to (3): （1）； （2）； （3）； In formulas (1) to (3), Indicates frequency point index, Let j represent the actual frequency of the k-th frequency point of the m-th frame signal, j represent the frequency channel index, n represent the n-th time node, and N represent the total number of time nodes of the m-th frame signal. This represents the signal value corresponding to the nth time node in the m-th frame of the signal, where i represents the imaginary unit. This represents the power response weight at the k-th frequency point on the j-th frequency channel of the m-th frame signal. Let represent the center frequency of the j-th filter in the dynamically compressed filter bank. This represents the equivalent rectangular bandwidth of the j-th filter. This represents the minimum tilt value of the dynamically compressed filter bank obtained by fitting the filter function. It is the arctangent function. This represents an exponential function with the natural constant e as its base.

4. The intelligent control method for a coal discharge hydraulic support according to claim 1, characterized in that, S3 includes: S31, calculate the dynamic tilt parameter of each frequency channel of each frame signal based on the power spectrum of each frequency channel of each frame signal; S32 inputs the dynamic tilt parameters of each frequency channel of each frame signal into the dynamically compressed filter bank to obtain the time-domain impulse response of each frequency channel of each frame signal.

5. The intelligent control method for a coal discharge hydraulic support according to claim 4, characterized in that, S31 is based on the power spectrum of the j-th frequency channel of the m-th frame signal. Calculate the dynamic tilt parameter of the j-th frequency channel of the m-th frame signal. When, this is achieved through formula (4): （4）； In formula (4), This represents the minimum tilt value of the dynamically compressed filter bank obtained by fitting the filter function. This represents the magnitude of the dynamic tilt parameter variation range, where S represents the tilt parameter of the dynamically compressed filter bank, and O represents the offset parameter of the dynamically compressed filter bank. This represents an exponential function with the natural constant e as its base.

6. The intelligent control method for a coal discharge hydraulic support according to claim 4, characterized in that, S32 dynamically tilts the j-th frequency channel of the m-th frame signal. The time-domain impulse response of the j-th frequency channel of the m-th frame signal is obtained by inputting it into the filter bank of the dynamic compression. When, this is achieved through formula (5): （5）； In formula (5), Let represent the center frequency of the j-th filter in the dynamically compressed filter bank. Let represent the equivalent rectangular bandwidth of the j-th filter, A represent the preset amplitude of the dynamically compressed filter bank, t represent time, q represent the order of the dynamically compressed filter bank, and b represent the preset attenuation factor. This represents the initial phase, and i represents the imaginary unit. Represents the logarithmic function. This represents an exponential function with the natural constant e as its base.

7. The intelligent control method for a coal discharge hydraulic support according to claim 1, characterized in that, S4 includes: The time-domain impulse response and its power spectrum of each frequency channel of each frame signal are weighted and summed in the frequency domain by using a frequency domain multiplication operation to obtain the compressed energy value of each frequency channel of each frame signal.

8. The intelligent control method for a coal discharge hydraulic support according to claim 1, characterized in that, The improved residual shrinkage model includes an input layer, a convolutional layer, multiple stacked residual shrinkage units, a global average pooling layer, and a fully connected classification layer. Each residual shrinkage unit includes a global average pooling unit, a fully connected unit, an activation unit, and an output unit. S5 includes: S51, calculate the Wiener entropy value of each frequency channel based on the compressed energy value of each frequency channel of all frame signals; S52, the spectrogram is input from the input layer and the feature is extracted through the convolutional layer to obtain the input features of each frame of signal; S53, through a multi-layer stacked residual shrinking unit, performs adaptive denoising on the input features of each frame signal in sequence, and the last residual shrinking unit outputs the denoised features of each frame signal. S54 converts the denoising features of each frame signal into feature vectors of each frame signal through a global average pooling layer, and outputs the recognition probability distribution of each frame signal by a fully connected classification layer. In this process, the stacked residual shrinking unit performs adaptive denoising on the input features of each frame of signal, while the global average pooling unit performs global average pooling on the input features of each frame of signal to obtain the pooling value of each frequency channel. Based on the pooling value of each frequency channel and the Wiener entropy value of each frequency channel, an enhancement vector for each frequency channel is constructed. The fully connected unit converts the enhancement vector of each frequency channel into a customized weight for each frequency channel. The activation unit calculates the frequency channel threshold based on the customized weight of each frequency channel and the pooling value of each frequency channel. The output unit performs adaptive denoising on the input features of each frame of signal using the frequency channel threshold.

9. The intelligent control method for a coal discharge hydraulic support according to claim 8, characterized in that, S51 includes: S511, calculate the arithmetic mean of the compressed energy values ​​of each frequency channel based on the sum of the compressed energy values ​​of each frequency channel of all frame signals; S512, calculate the geometric mean of the compressed energy value of each frequency channel based on the sum of the logarithms of the compressed energy values ​​of each frequency channel of all frame signals; S513, calculate the ratio of the geometric mean to the arithmetic mean of the compressed energy values ​​for each frequency channel to obtain the Wiener entropy value for each frequency channel.

10. The intelligent control method for a coal discharge hydraulic support according to claim 1, characterized in that, S6 includes: S61, determine whether the probability of coal release in the recognition probability distribution result of any frame signal is greater than the preset threshold. If the probability of coal release in the recognition probability distribution result of any frame signal is greater than the preset threshold and the probability of gangue release in the recognition probability distribution results of multiple consecutive frames of signals is less than the preset threshold, then determine that the current state of the fully mechanized mining face is coal release. S62, determine whether the probability of waste rock release in the recognition probability distribution results of consecutive multi-frame signals is greater than a preset threshold. If the probability of waste rock release in the recognition probability distribution results of consecutive multi-frame signals is greater than the preset threshold, then determine that the current state of the fully mechanized mining face is waste rock release. S63 generates a control strategy for the current state of the coal release hydraulic support based on the current state of the fully mechanized coal release face, and controls the coal release hydraulic support to execute the control strategy.

Images